R3b [fw]: delete legacy workflow-script/v1 composite and cp-service-token/v1 context (dbc5efc6)

Control plane 0.0.5.1890 no longer falls back to either. Invariant tests now
assert both stay gone.

Claude-Session: https://claude.ai/code/session_01N218bERyNeLhbLmDRDvbWF

CLAUDE.md

@@ -49,8 +49,12 @@ task-templates/          # dispatch-task --template <name> YAML files
 workflows/               # Workflow DAG definitions
 model-registry/          # Model scoring YAMLs (capability dimensions)
 models/                  # Model endpoint YAMLs (URLs, auth, parameters)
+benchmarks/              # Reference data for the scores/costs pass (NOT read by any loader):
+                         #   artificial-analysis/ snapshot + README, unverified/estimate files
 ```
 
+**Model slugs:** a registry slug is not a provider model id. MiniMax silently maps unknown model names to MiniMax-M3, so each `models/<slug>.yaml` must carry the real id in `default_flags.model` (agent-runtimes translates slug -> id on every runner). One `models/` + `model-registry/` file per effort level (e.g. `minimax-m31-flash-{max,medium}`).
+
 ## Harness Conventions
 
 ### harness.yaml fields

CONTEXT.md

@@ -0,0 +1,7 @@
+# CONTEXT.md — agent-runtime-framework
+
+## Active
+
+- **MiniMax Flash migration done; scores/costs are placeholders.** `minimax-m31-flash-{max,medium}` (models/ + model-registry/) and context `minimax-m31-flash/v1` are live on the CP (d7decc1). Registry scores are copies of the M3 entries with costs just below them. Planner FUTURE `629829d3` (agent-runtimes) tracks replacing them and the stop-gap `benchmarks/estimates.csv` M3.1-Flash row (0.9x M3 cost, index 61.5 — implausible). See `benchmarks/artificial-analysis/README.md` for the AA snapshot, slug mapping and caveats.
+- **Unverified end-to-end:** dispatch a real hatchmate task and check the model the API reports is `MiniMax-M3.1-Flash-Preview` (agent-runtimes 0.0.5.1817 translates slug -> provider id on all runners).
+- Untracked, left alone: `.claude/` (worktrees), `tests/__init__.py`.

benchmarks/artificial-analysis/2026-10-13.csv

@@ -0,0 +1,35 @@
+model,creator,context,intelligence_index,cost_per_task_usd,cost_per_intelligence_point_usd,median_tokens_per_s
+Claude Opus 5.5 (max with fallback),Anthropic,1M,58,5.98,0.1031,96
+Claude Sonnet 5.5 (max with fallback),Anthropic,1M,56,5.46,0.0975,135
+Claude Opus 5.5 (xhigh with fallback),Anthropic,1M,56,3.46,0.0618,81
+Claude Opus 5.5 (high with fallback),Anthropic,1M,54,1.82,0.0337,82
+Claude Fable 5.1 (max with fallback),Anthropic,1M,53,7.63,0.1440,67
+Claude Fable 5.1 (xhigh with fallback),Anthropic,1M,53,5.98,0.1128,62
+Claude Sonnet 5.5 (xhigh with fallback),Anthropic,1M,52,2.01,0.0387,103
+Claude Opus 5.5 (medium with fallback),Anthropic,1M,51,1.34,0.0263,79
+Claude Fable 5.1 (high with fallback),Anthropic,1M,51,3.91,0.0767,52
+Claude Fable 5.1 (medium with fallback),Anthropic,1M,49,2.98,0.0608,54
+Claude Fable 5.1 (low with fallback),Anthropic,1M,47,2.37,0.0504,56
+Claude Sonnet 5.5 (high with fallback),Anthropic,1M,47,0.88,0.0187,96
+Claude Haiku 5.5 (max),Anthropic,1M,43,0.21,0.0049,236
+Claude Opus 5.5 (low with fallback),Anthropic,1M,42,0.55,0.0131,80
+Claude Haiku 5.5 (xhigh),Anthropic,1M,41,0.12,0.0029,187
+Claude Sonnet 5.5 (medium with fallback),Anthropic,1M,41,0.48,0.0117,95
+Claude Haiku 5.5 (high),Anthropic,1M,38,0.08,0.0021,175
+Claude Sonnet 5.5 (low with fallback),Anthropic,1M,36,0.35,0.0097,100
+Claude Haiku 5.5 (medium),Anthropic,1M,34,0.05,0.0015,171
+Claude Haiku 5.5 (low),Anthropic,1M,29,0.02,0.0007,164
+MiniMax-M3,MiniMax,1M,29,0.51,0.0176,107
+Qwen3.8 Max (0902),Alibaba,984k,45,5.41,0.1202,36
+Qwen3.8 2.4T A95B,Alibaba,262k,40,2.16,0.0540,37
+Qwen3.8-Flash-Next,Alibaba,256k,40,0.37,0.0092,51
+Qwen3.8 27B (xhigh),Alibaba,256k,34,1.01,0.0297,46
+Qwen3.8 27B (medium),Alibaba,256k,28,1.13,0.0404,50
+Qwen3.8 27B (low),Alibaba,256k,26,1.05,0.0404,53
+Qwen3.7 Plus,Alibaba,1M,25,0.22,0.0088,55
+Qwen3.8 27B (non-reasoning),Alibaba,256k,20,2.49,0.1245,52
+Qwen3.5 397B A17B,Alibaba,262k,18,0.47,0.0261,88
+Qwen3.6 35B A3B,Alibaba,262k,18,0.48,0.0267,129
+Qwen3.5 122B A10B,Alibaba,262k,16,0.32,0.0200,129
+Qwen3.5 9B,Alibaba,262k,11,0.21,0.0191,58
+Qwen3 Coder Next,Alibaba,256k,9,0.78,0.0867,73

benchmarks/artificial-analysis/README.md

@@ -0,0 +1,28 @@
+# Artificial Analysis benchmark snapshot
+
+Reference data for the model-registry scores/costs pass. NOT read by any loader
+(RegistryEntry ignores unknown fields, so this lives beside the registry, not in it).
+
+- **Source:** https://artificialanalysis.ai/leaderboards/models, extracted from the rendered
+  table on 2026-10-13 (Anthropic, Alibaba/Qwen and MiniMax rows only; 34 rows with a score and a cost).
+- **File:** `2026-10-13.csv` — `intelligence_index` is the AA Intelligence Index; `cost_per_task_usd` is the
+  site's "Cost per Task (USD)".
+- **`cost_per_intelligence_point_usd`** = `cost_per_task_usd / intelligence_index`. Derived here, not a site
+  column. Chosen as the cost factor to record (user decision, 2026-10-13).
+- Rows the site marks with `*` (partial results, no cost shown) are omitted.
+- Names are the site's labels. "with fallback" on most Claude rows is the site's wording; effort level is the
+  bracketed suffix.
+- Attribution: data © Artificial Analysis. Re-extract rather than editing by hand.
+
+## Mapping to registry slugs (informational, not authoritative)
+
+| Registry slug | AA row(s) | Note |
+|---|---|---|
+| `minimax-m3`, `minimax-m3-max` | MiniMax-M3 (index 29, $0.51, 0.0176) | AA lists one M3 row with no effort level; both slugs share it |
+| `minimax-m31-flash-{max,medium}` | none (see `../estimates.csv`) | MiniMax-M3.1-Flash-Preview is NOT on the leaderboard yet (checked 2026-10-13; no model page). **TODO(revisit):** `../estimates.csv` holds a placeholder estimate (cost 0.9x M3, index scaled by the KingBench ratio) — replace with measured values |
+| `airouter-qwen38`, `airouter-qwen38-max` | Qwen3.8 family | `default_flags.model: "Qwen3.8"` does not say which size; AA has Qwen3.8 27B (low/medium/xhigh), 2.4T A95B, Max (0902), Flash-Next — confirm before mapping |
+| `claude-fable-5-1` | Claude Fable 5.1 (low..max) | registry sets `reasoning_effort: low` (inert on the claude runner) |
+| `claude-sonnet-5`, `claude-opus-5`, `claude-haiku-4-5` | Sonnet 5.5, Opus 5.5, Haiku 5.5 | slug/model-id names predate the 5.5 releases (`default_flags.model` is `claude-sonnet-5` etc.); verify which API model each actually calls before using these rows |
+
+Cost per task is not comparable to `cost_per_1k_tokens`: it folds in how many tokens the model spends. Use it
+for `cost_efficiency` scoring, not for the per-token price fields.

benchmarks/estimates.csv

@@ -0,0 +1,2 @@
+model,creator,context,intelligence_index,cost_per_task_usd,cost_per_intelligence_point_usd,status,basis
+MiniMax-M3.1-Flash-Preview (effort unspecified),MiniMax,1M,61.5,0.46,0.0075,ESTIMATE-TODO,"cost = 0.9 x MiniMax-M3 AA cost per task ($0.51); index = M3 AA index (29) x KingBench-3 ratio (66.25/31.25 = 2.12); both inputs unverified/placeholder"

benchmarks/minimax-m3-1-flash-unverified.md

@@ -0,0 +1,23 @@
+# MiniMax-M3.1-Flash-Preview — ad-hoc, UNVERIFIED figures
+
+Supplied by the operator on 2026-10-13 with no source URL. Not from Artificial Analysis (the model is not on
+that leaderboard yet — see `artificial-analysis/README.md`). Do not use for registry scores until confirmed.
+
+| Item | Figure | Note |
+|---|---|---|
+| KingBench 3, M3.1 Flash | 66.25% (53/80) | "community and third-party testing"; arithmetic checks (53/80 = 66.25%) |
+| KingBench 3, M3 baseline | 31.25% (25/80) | arithmetic checks; the source also mentions "alternative baseline figures depending on the evaluation set" |
+| M3 SWE-bench Pro | ~59% | reference baseline for M3, described as official |
+| M3 "Browse.com" | 83.5 | benchmark name looks mistyped (BrowseComp?) — confirm before citing |
+| M3 Terminal Bench 2.1 | "strong" | no number given |
+
+Caveats: unknown effort level for the KingBench runs (Flash always thinks; server default is max); KingBench 3 is
+small (80 items) so one item is 1.25 points; none of these figures share a scale with the AA Intelligence Index
+(M3 = 29 there), so they cannot be mixed into one cost-per-intelligence figure.
+
+## TODO(revisit) — placeholder estimate in use (2026-10-13)
+`estimates.csv` carries a stop-gap row for M3.1 Flash: cost per task = 90% of M3's ($0.51 -> $0.46) and
+intelligence index = M3's AA index (29) scaled by the KingBench-3 ratio (66.25/31.25 = 2.12) = 61.5. That index
+would rank it above Claude Opus 5.5 max (58), which is implausible for a Flash-class model — treat both numbers as
+rough placeholders. Replace with (a) the Artificial Analysis row when the model is listed, or (b) our own measured
+cost per task at each effort level (fixed task set via `ask-minimax-run --output-format json`).

harnesses/composites/best-practices-minimax-repo/v1/harness.yaml

@@ -6,6 +6,6 @@ description: "Best practices review agent on MiniMax M3 (max effort) + repo clon
 layers:
   - context: best-practices/v1
   - context: minimax/v1
-  - context: minimax-m3-max/v1
+  - context: minimax-m31-flash/v1
   - context: gitea-ssh/v1
   - context: agent-repo/v1

harnesses/composites/integration-direct/v1/harness.yaml

@@ -9,7 +9,7 @@ layers:
   # personal-subscription OAuth — integration runs are frequent/mechanical
   # and were draining the interactive 5-hour window (2026-09-02).
   - context: minimax/v1
-  - context: minimax-m3-max/v1
+  - context: minimax-m31-flash/v1
   - context: gitea-ssh/v1
   - context: agent-repo/v1
   - context: direct-push/v1

harnesses/composites/planning-sonnet-repo/v1/harness.yaml

@@ -8,8 +8,3 @@ layers:
   - context: anthropic-cloud-paul-oauth/v1
   - context: gitea-ssh/v1
   - context: agent-repo/v1
-  # AU-51/AU-51b: this composite backs bug-verify-sonnet@1, whose agent node
-  # self-transitions its own item's flow_state (PATCH /v1/work-items/{id})
-  # as its final step — see work item 88ecd23e ("agent self-transition"
-  # pattern, spec/typed-workflows.md). Needs a CP write credential.
-  - context: cp-service-token/v1

harnesses/composites/scaffolding-repo/v1/harness.yaml

@@ -8,6 +8,6 @@ layers:
   # MiniMax M3 — scaffolding is mechanical (stub files so tests import);
   # moved off the personal Anthropic subscription 2026-09-02.
   - context: minimax/v1
-  - context: minimax-m3-max/v1
+  - context: minimax-m31-flash/v1
   - context: gitea-ssh/v1
   - context: agent-repo/v1

harnesses/composites/spec-writing-sonnet-repo/v1/harness.yaml

@@ -8,6 +8,3 @@ layers:
   - context: anthropic-cloud-paul-oauth/v1
   - context: gitea-ssh/v1
   - context: agent-repo/v1
-  # AU-51/AU-51b: this composite backs scope-decompose-sonnet@1, which
-  # calls back POST /v1/work-items/{id}/decompose — needs a credential.
-  - context: cp-service-token/v1

harnesses/composites/test-writing-repo/v1/harness.yaml

@@ -9,6 +9,6 @@ layers:
   # 2026-09-02 (the "credential layer swappable" swap this composite was
   # designed for).
   - context: minimax/v1
-  - context: minimax-m3-max/v1
+  - context: minimax-m31-flash/v1
   - context: gitea-ssh/v1
   - context: agent-repo/v1

harnesses/composites/workflow-script-nocp-norepo/v1/harness.yaml

@@ -0,0 +1,14 @@
+kind: composite
+name: workflow-script-nocp-norepo
+version: 1
+description: "Internal composite for kind: script nodes with cp_write=false and NO effective agent_repo (typed-workflows WF-COMP-31e, planner item e993c9fa). Brings only the shared-variables filesystem context: no repo-readonly clone, no CP credential, and therefore ZERO secrets_required entries. That makes the node eligible for an M45 tenant namespace (the NS-12 gate rejects any ESO account secret) and for the M33 egress broker sidecar, which supplies the node's credentials through bindings. Not model-bearing; the compiled shape is cli=\"shell\", params.cmd=<cmd>."
+
+# Counterpart of workflow-script-nocp/v1 for templates that declare no agent_repo.
+# The compiler (controlplane/script_composites.py resolve_script_composite) selects
+# this only when effective_repo is False; with AGENT_REPO_URL unset the entrypoint
+# V2 guard does not require /workspace/project. Never layer repo-readonly/v1 here:
+# its gitea-ssh-readonly secret is exactly what this composite exists to avoid.
+layers:
+  # spec/workflow-variables.md WV-19 / spec/harness.md HD-14 — provides the
+  # [shared-variables] tag.
+  - context: shared-variables/v1

harnesses/composites/workflow-script/v1/harness.yaml

@@ -1,35 +0,0 @@
-kind: composite
-name: workflow-script
-version: 1
-description: "Internal composite for the flag-on kind: script path (typed-workflows WF-COMP-31 / workflow-variables WV-19). Brings the shared-variables filesystem context AND the AU-51 CP write credential (cp-service-token/v1 sets CP_SERVICE_TOKEN_FILE in the entrypoint process env before shell_runner runs). NOT a model-bearing composite — the kind: script compiled shape is cli=\"shell\", params.cmd=<cmd>; this composite provides the harness layers that path needs and nothing else."
-
-# Order matters only for HC-6 env last-writer-wins / HC-8 git identity. The two
-# layers here have disjoint env vars (shared-variables sets WORKFLOW_VARIABLES_*
-# paths, cp-service-token sets CP_SERVICE_TOKEN_FILE) and no overlapping git
-# identity, so the listed order has no observable merge effect — both layers
-# reach the script-node container either way.
-layers:
-  # spec/workflow-variables.md WV-19 / spec/harness.md HD-14 — agent-facing
-  # /variables/ context (CLAUDE.md mounted at /opt/harness/context/shared-variables/CLAUDE.md)
-  # plus WORKFLOW_VARIABLES_ROOT=/variables and templated WORKFLOW_VARIABLES_SELF=<step_path>.
-  # Provides the [shared-variables] tag used by the dispatcher's extra_layers injection.
-  - context: shared-variables/v1
-  # spec/typed-workflows.md WF-COMP-31 "AU-51 credential env seam" — this layer's
-  # init.sh restages the ESO-mounted CP-service-token secret (root-owned mode 0400)
-  # to an agent-owned 0600 copy at /run/agent/cp-service-token/token and exports
-  # CP_SERVICE_TOKEN_FILE pointing at it. Verified: env.CP_SERVICE_TOKEN_FILE is set
-  # on the entrypoint process environment before shell_runner runs (entrypoint reads
-  # os.environ via env=os.environ.copy() in subprocess.run at entrypoint/runners/shell.py:73,
-  # so the same credential channel an agent node gets reaches a script node unchanged).
-  - context: cp-service-token/v1
-  # spec/typed-workflows.md WF-COMP-34 (bug 3bee78fa, 2026-10-02) — read-only
-  # project-repo clone for harness-path kind: script nodes. Layered LAST so
-  # its clone runs after the two AU-51 wiring layers and before the script
-  # starts (init.sh order: shared-variables -> cp-service-token -> repo-readonly).
-  # When AGENT_REPO_URL is unset (the pre-fix runtime, or any template that
-  # does not declare agent_repo) this layer is a no-op; the V2 guard
-  # (entrypoint/entrypoint.py:901-906) does NOT trip because the unset case
-  # is a no-op there too. Layered-in instead of layering agent-repo/v1
-  # because that context's fork management and push side effects are wrong
-  # for a read-only script node (see bug item body).
-  - context: repo-readonly/v1

harnesses/contexts/agent-repo/v1/init.sh

@@ -65,6 +65,41 @@ if [ -n "${AGENT_REPO_URL:-}" ] && [ -n "${AGENT_BRANCH:-}" ]; then
     echo "Cloning agent working repo: $AGENT_REPO_URL (branch: $AGENT_BRANCH)"
     if git clone --depth 1 --branch "$AGENT_BRANCH" "$AGENT_REPO_URL" /workspace/project 2>/dev/null; then
         echo "Cloned existing branch $AGENT_BRANCH (continuing prior work)"
+        # AR-42: archive the stale task branch and reset it to the current base.
+        # Opt-in per node (AGENT_BRANCH_RESET_ON_START=true, set by the workflow
+        # template on its FIRST node only): every node of a workflow shares this
+        # one branch, so an unconditional reset would wipe earlier nodes' commits.
+        if [ "${AGENT_BRANCH_RESET_ON_START:-}" = "true" ]; then
+            mkdir -p /workspace/.agent-output
+            AR42_TOOL=/opt/agent/agent_repo_base.py
+            AR42_FAIL() {
+                echo "ERROR: AR-42 $1" >&2
+                printf '{"error": "ar42_archive_and_reset_failed", "detail": %s}\n' \
+                    "$(printf '%s' "$1" | python3 -c 'import json,sys; print(json.dumps(sys.stdin.read()[:500]))')" \
+                    > /workspace/.agent-output/init-error.json
+                exit 1
+            }
+            AR42_SEL=$(python3 "$AR42_TOOL" select-base \
+                --workdir /workspace/project \
+                --branch "$AGENT_BRANCH" \
+                --upstream-url "${AGENT_UPSTREAM_REPO_URL:-}" \
+                --upstream-branch "${AGENT_UPSTREAM_BRANCH:-main}" \
+                --reference-dir /workspace/reference/main \
+                --out /workspace/.agent-output/branch_base.json 2>&1) \
+                || AR42_FAIL "select-base failed: $AR42_SEL"
+            AR42_SHA=$(printf '%s\n' "$AR42_SEL" | sed -n 's/^sha=//p' | head -1)
+            [ -n "$AR42_SHA" ] || AR42_FAIL "select-base returned no sha: $AR42_SEL"
+            AR42_OUT=$(python3 "$AR42_TOOL" archive-and-reset \
+                --workdir /workspace/project \
+                --branch "$AGENT_BRANCH" \
+                --base-sha "$AR42_SHA" \
+                --fork-url "$AGENT_REPO_URL" \
+                --archive-prefix archive \
+                --base-metadata /workspace/.agent-output/branch_base.json \
+                --out /workspace/.agent-output/branch_base.json 2>&1) \
+                || AR42_FAIL "archive-and-reset failed: $AR42_OUT"
+            echo "AR-42: $AGENT_BRANCH reset to base $AR42_SHA"
+        fi
     else
         echo "Branch $AGENT_BRANCH does not exist — seeding fresh branch from upstream reference"
         # Full clone (not --depth 1) so we get a working remote for finalize.sh push.

harnesses/contexts/cp-docs-token/v1/CLAUDE.md

@@ -0,0 +1,48 @@
+# Step documents (cp-docs-token)
+
+This step can read and write planner documents through `python3 -m lib.cp_cli`
+in step-docs mode (PLN-CLI-16). The mode is active because `CP_DOCS_TOKEN_FILE`
+and `CP_DOCS_CA_FILE` are set and the listener URL is in `CP_DOCS_URL`.
+
+## Credential
+
+- The token is a file at `$CP_DOCS_TOKEN_FILE`. cp-cli reads it on every call.
+- Never print, copy, log or pass the token on a command line.
+- Do not use `~/.cpconfig`; step-docs mode does not read it.
+
+## Allowed commands
+
+Only these commands work; anything else exits 2 `cp_docs_mode_command_not_allowed`:
+
+```
+python3 -m lib.cp_cli planner items documents list <item-uuid>
+python3 -m lib.cp_cli planner items documents get <item-uuid> <name> [--json]
+python3 -m lib.cp_cli planner items documents put <item-uuid> <name> --file <path> (--create-only | --if-match <rev>)
+python3 -m lib.cp_cli planner items documents related [--json]
+```
+
+Item arguments are full UUIDs. Do not pass a project id.
+
+## Scope
+
+- You may write only documents on this step's own item.
+- You may read documents on this step's item, its parent and its siblings.
+- `related` prints `item_uuid`, `parent_uuid`, `sibling_uuids` and `siblings_truncated`.
+
+## Writes need a precondition
+
+Exactly one of these is required on every `put` (neither or both exits 2):
+
+- `--create-only` creates a new document (sends `If-None-Match: *`).
+- `--if-match <rev>` updates an existing one (sends `If-Match`). Take `<rev>`
+  from `get --json` (the ETag revision); surrounding quotes are stripped.
+
+## Errors
+
+- 412 `precondition_failed`: the document changed or already exists. Re-run
+  `get --json`, merge your change into the current content, and `put` again
+  with the new revision.
+- 428 (precondition missing): add `--create-only` or `--if-match`.
+- 429 `step_docs_write_budget_exhausted` or `step_docs_rate_limited`: stop
+  writing; do not retry in a loop.
+- 404 `not_found`, 400 `invalid_document_name`: fix the item uuid or name.

harnesses/contexts/cp-docs-token/v1/harness.yaml

@@ -0,0 +1,16 @@
+kind: context
+name: cp-docs-token
+version: 1
+description: "Step-docs mode env contract for lib.cp_cli (PLN-CLI-16): per-step wfcd_ token file and CA file delivered by the dispatcher; document access scoped to this step's item (spec/dispatcher.md D-40, spec/authentication.md AU-59)"
+requires: []
+provides: [cp-docs-token]
+
+# Env-only layer: the dispatcher delivers the token and CA files into the
+# container; this layer sets only the paths. No secrets_required, no init script.
+env:
+  CP_DOCS_TOKEN_FILE: /run/agent/cp-docs/token
+  CP_DOCS_CA_FILE: /run/agent/cp-docs/ca.crt
+
+context_files:
+  - source: ./CLAUDE.md
+    target: /opt/harness/context/cp-docs-token/CLAUDE.md

harnesses/contexts/cp-service-token/v1/harness.yaml

@@ -1,30 +0,0 @@
-kind: context
-name: cp-service-token
-version: 1
-description: "CP-internal service bearer token — authenticates agent-container write calls back to the control plane (AU-51/AU-51b, bug b3a96acc)"
-requires: []
-provides: []
-
-# AU-51b: the ESO-mounted secret is root-owned mode 0400 (init.sh runs as
-# root, before the gosu drop to the agent user, uid 1000) — unreadable by
-# the agent process directly. Confirmed live 2026-09-03: a real agent
-# session hit exactly this ("cp-cli invocation can't proceed... token is
-# root-owned mode 0400 so it's not readable by the agent user"). Mirrors
-# anthropic-cloud-paul-oauth/v1's pattern: init.sh restages a 0600
-# agent-owned copy outside the ESO mount. CP_SERVICE_TOKEN_FILE points at
-# the restaged copy, not the raw mount.
-#
-# The value at this path MUST match the CP's CP_INTERNAL_BEARER_TOKEN
-# (same account: cp-decompose-service-token, account_id
-# 1d963673-6ac9-4f85-875a-2ce5323e76ad, owner (cp, cp)).
-env:
-  CP_SERVICE_TOKEN_FILE: /run/agent/cp-service-token/token
-
-scripts:
-  init: "./init.sh"
-
-secrets_required:
-  - name: cp-service-token
-    account_ref: "1d963673-6ac9-4f85-875a-2ce5323e76ad"
-    mount_path: /run/agent/secrets/cp-service-token
-    mode: "0400"

harnesses/contexts/cp-service-token/v1/init.sh

@@ -1,39 +0,0 @@
-#!/bin/bash
-# cp-service-token init — restage the ESO-mounted secret to an agent-owned
-# copy.
-#
-# Threat model: keep the ESO mount root-only (mode 0400) so the agent user
-# cannot directly `cat` /run/agent/secrets/cp-service-token/*. init.sh runs
-# as root (in uid-wrapper.sh, before the gosu drop) and writes a restaged
-# copy at /run/agent/cp-service-token/token with mode 0600 owned by the
-# agent user. This mirrors anthropic-cloud-paul-oauth/v1/init.sh's pattern.
-#
-# Bug discovered live 2026-09-03: without this restaging step, cp-cli
-# (invoked by the agent user, uid 1000) could not read the raw 0400
-# root-owned mount at all — "cp_cli invocation can't proceed... token is
-# root-owned mode 0400 so it's not readable by the agent user". The context
-# originally set CP_SERVICE_TOKEN_FILE directly at the raw ESO mount path,
-# which is the mistake this init.sh fixes.
-
-set -euo pipefail
-
-ESO_TOKEN="/run/agent/secrets/cp-service-token/value"
-
-if [ ! -r "$ESO_TOKEN" ]; then
-    echo "ERROR: $ESO_TOKEN not readable. Check ESO ExternalSecret acct-1d963673." >&2
-    exit 1
-fi
-
-AGENT_USER="${AGENT_USER:-agent}"
-
-DEST_DIR="/run/agent/cp-service-token"
-mkdir -p "$DEST_DIR"
-
-DEST_FILE="$DEST_DIR/token"
-cp "$ESO_TOKEN" "$DEST_FILE"
-
-chown "$AGENT_USER:" "$DEST_DIR" "$DEST_FILE" 2>/dev/null || true
-chmod 0700 "$DEST_DIR"
-chmod 0600 "$DEST_FILE"
-
-echo "cp-service-token: restaged into $DEST_FILE (0600 $AGENT_USER:$AGENT_USER)"

harnesses/contexts/minimax-m31-flash/v1/harness.yaml

@@ -0,0 +1,17 @@
+kind: context
+name: minimax-m31-flash
+version: 1
+description: "Pin MiniMax-M3.1-Flash-Preview (1M context) — layer AFTER minimax/v1 (auth + base URL come from there). Thinking depth is set per dispatch via --effort (models/*.yaml default_flags.reasoning_effort), not here."
+requires: []
+provides: []
+
+# Replaces minimax-m3-max/v1 (kept for rollback). Differences:
+# - Flash always thinks and ignores MAX_THINKING_TOKENS/budget_tokens (effort is
+#   the only depth control, server default max), so there is no init.sh.
+# - No [1m] suffix: Flash is 1M natively (verified 2026-10-13, modelUsage.contextWindow).
+# - CLAUDE_CODE_AUTO_COMPACT_WINDOW stays: an unknown model id otherwise gets
+#   Claude Code's 200K default and compacts early.
+env:
+  ANTHROPIC_MODEL: "MiniMax-M3.1-Flash-Preview"
+  ANTHROPIC_SMALL_FAST_MODEL: "MiniMax-M3.1-Flash-Preview"
+  CLAUDE_CODE_AUTO_COMPACT_WINDOW: "1000000"

memory/log/2026-05-04.225112.md

@@ -1,27 +0,0 @@
-# Session Log -- 2026-05-04
-
-## Summary
-Implemented 5 planned fixes for BUG-5 (`.agent-output/` silent drops) and BUG-20 (MiniMax finalize no-push): pre_test.sh harness hook, output validation in finalize, push retry with debug tracing, spec-planning workflow output tags, and wrong-path detection with agent correction re-invoke. Fork cleanup PR merged, agent-monitor PR #44 merged.
-
-## Decisions
-- Fork cleanup workflow (PR #1 on agent-runtimes-agents): weekly cron deletes `task-*` branches inactive for 7+ days; main reset to orphan commit if no commits for 7+ days. Supports dry-run via workflow_dispatch. **Merged** (2026-05-04 10:39 UTC by admin).
-- Gitea API token auth requires HTTP Basic Auth (`auth=("user","pass")`) not Bearer token — Bearer returns 401 even with valid access token. Use `httpx.get(..., auth=("user","pass"))` pattern.
-- Shallow clone (`--depth 1`) for task branches in fork cleanup to avoid bloating with full history.
-
-## Gotchas Discovered
-- **[Gitea Gotchas]** Gitea API endpoint `/login` is `404 Not Found` — use `/users/<username>/access_tokens` to create tokens, or use HTTP Basic Auth directly against any `/api/v1/` endpoint with `auth=("user","pass")`.
-- **[Gitea Gotchas]** PR state `closed` with `merged: true` means the PR was merged (not just closed). Gitea distinguishes merged vs closed states.
-- **[Gitea Gotchas]** Fork PRs use the parent repo's Actions (has_actions=true on parent, `has_actions: false` on fork). Fork cleanup workflow runs on the fork's GHA because fork enables Actions independently.
-- **[tokens]** Token file at `~/.config/agent-runtimes/tokens.json` uses flat structure: `{"access_token": "...", "expires_at": N}`. Token was valid (not expired at session continuation time: now 1777891716 vs expiry 1777891804).
-
-## Key Context
-- **agent-runtime-framework** `8bbf6cb`: pre_test.sh created (harness-level test folder revert), finalize.sh gains `set -x` tracing, push retry (`AGENT_PUSH_RETRIES`), output validation (`AGENT_EXPECTED_OUTPUT`), wrong-path detection + `.correction-prompt.txt` sentinel for entrypoint re-invoke. All 17 spec-planning workflow nodes tagged with `output: {path, min_bytes}`.
-- **agent-runtimes-agents** fork: `fork-cleanup` branch merged to main (PR #1 by ai_enablement, merged by admin 2026-05-04). Workflow file at `.gitea/workflows/cleanup.yaml`.
-- **agent-runtimes** `148d9d3`: PR #44 merged — `scripts/agent-monitor` gains `provider` and `model_full` columns.
-- **agent-runtimes** local uncommitted: M22 Phase 3 manifests work (`controlplane/api/manifests.py`, `controlplane/db/manifest_store.py`, migration `20260504_0017_m16_phase3_manifests.py`, `controlplane/manifests/`). Not related to this session's focus.
-- Two ops gates pending before M22 Phase 9 live cutover: CI build `agent-runtimes-init:1.0.0` image; `agent-session` Role deployed via `homelab/agent-runtimes-deploy` + ArgoCD sync.
-
-## Process Notes
-- When verifying PR state via Gitea API: parse JSON directly (don't assume specific keys — use `json.dumps(..., indent=2)` to inspect full structure).
-- Fork cleanup workflow commits to `agent-runtimes-agents` fork main; parent `agent-runtimes` main unchanged. Fork's main tracked via `agent-runtimes-agents` remote in local `agent-runtime-framework` checkout.
-- Fork cleanup's `fetch-depth: 0` is needed for commit date inspection even with shallow clones elsewhere.

memory/log/2026-10-13.144722.md

@@ -0,0 +1,32 @@
+# Session Log -- 2026-10-13
+
+## Summary
+Moved the framework's MiniMax content to MiniMax-M3.1-Flash-Preview (new context, Flash models/registry, bare `minimax`/`minimax-m2` slugs removed; d7decc1), then added a `benchmarks/` directory with an Artificial Analysis snapshot, unverified M3.1 Flash figures, and a flagged placeholder estimate (418d634). Both pushed to `main`.
+
+## Decisions
+- Decision: new context `harnesses/contexts/minimax-m31-flash/v1` (env pin `MiniMax-M3.1-Flash-Preview`, no init.sh) replaces `minimax-m3-max/v1` in scaffolding-repo, integration-direct, test-writing-repo, best-practices-minimax-repo; the old context stays for rollback -- Rationale: Flash ignores MAX_THINKING_TOKENS; effort comes from `--effort` via `default_flags.reasoning_effort`.
+- Decision: deleted `model-registry/minimax.yaml` (endpoint `minimax` never existed) and repointed the six `task-templates/minimax-*.yaml` and the three `*-review-loop@1` `model: minimax-m2` nodes to `minimax-m31-flash-max` -- Rationale: MiniMax silently maps unknown model names to M3.
+- Decision: benchmark data lives in `benchmarks/` beside the registry, not inside registry entries -- Rationale: `RegistryEntry` (pydantic default extra=ignore) silently drops unknown fields, and the loader never reads this dir.
+- Decision: cost factor recorded is "cost per intelligence point" = AA cost per task / AA Intelligence Index (derived; the site has no such column) -- Rationale: operator request.
+- Decision: M3.1 Flash placeholder = 0.9x M3 cost per task ($0.46) and M3's index scaled by the KingBench-3 ratio (29 x 2.12 = 61.5), flagged `ESTIMATE-TODO` -- Rationale: operator stop-gap; 61.5 exceeds Opus 5.5 max (58), so implausible. Planner FUTURE item 629829d3 (agent-runtimes) tracks replacing it.
+
+## Gotchas Discovered
+- **[minimax]** Symptom: unknown model names (`minimax`, `minimax-m31-flash-max`) return HTTP 200 with `model: MiniMax-M3` -- Fix: send the provider id; translation now done by agent-runtimes ef822703 (deployed 0.0.5.1817). Check the echoed `model`.
+- **[artificial-analysis]** Symptom: WebFetch of artificialanalysis.ai/models returns only headline numbers (tables are JS-rendered) -- Fix: open the page in Chrome (claude-in-chrome), wait ~4 s, read `document.querySelector('table')` rows via javascript_tool (269 rows on /leaderboards/models; columns: model, context, creator, index, cost per task, tokens/s, latency).
+- **[artificial-analysis]** MiniMax-M3.1-Flash-Preview is not on the leaderboard (no row; /models/minimax-m3-1-* 404 as of 2026-10-13); only `MiniMax-M3` (index 29, $0.51/task) exists for MiniMax.
+- **[benchmarks]** Cost vs intelligence correlates strongly only within one model family across effort levels (Claude: r 0.93-0.99, cost x4-6 per +10 index points); across vendors it is weak (r 0.53) and Qwen3.8 27B is inverted (cost does not fall with lower effort), so cross-vendor cost estimates for M3.1 Flash are unreliable.
+- **[git]** The framework checkout was on the user's feature branch `feat/hatchmate-per-project-fork-key`; moving my uncommitted work to `main` needed `git stash push -- <paths>` (untracked new files stay in place), `checkout main`, ff-pull, `stash pop`.
+
+## Open Questions
+- Which Qwen3.8 size do `airouter-qwen38*` models actually call? (AA has 27B, 2.4T A95B, Max, Flash-Next.)
+- Registry model ids `claude-sonnet-5` / `claude-opus-5` / `claude-haiku-4-5` predate the 5.5 releases AA lists -- verify before mapping AA rows to them.
+- Source for the KingBench 3 figures; scoring rule to map AA index / cost per intelligence point onto the registry's 0-10 dimensions.
+
+## Key Context
+- Files: `benchmarks/artificial-analysis/{2026-10-13.csv,README.md}` (34 rows: Anthropic, Qwen, MiniMax-M3), `benchmarks/minimax-m3-1-flash-unverified.md`, `benchmarks/estimates.csv`.
+- Live: CRS checkout for this repo is at d7decc1 on the controlplane (verified); 418d634 adds only benchmarks (nothing reads it).
+- Registry scores/costs for `minimax-m31-flash-{max,medium}` are still placeholders (copies of the M3 entries; costs just under them).
+
+## Process Notes
+- Before rebasing onto a feature-branch checkout, check `git branch --show-current`; land work on `main` explicitly.
+- Record derived metrics with their derivation (README) and mark stop-gap numbers with a TODO in the data file name/status column AND a durable planner item.

model-registry/harness-roles.yaml

@@ -5,5 +5,6 @@ roles:
     claude-sonnet-5: code-sonnet-tdd-repo
     claude-opus-5: code-opus-tdd-repo
     claude-haiku-4-5: code-haiku-tdd-repo
-    minimax: code-minimax-tdd-repo
+    minimax-m31-flash-max: code-minimax-tdd-repo
+    minimax-m31-flash-medium: code-minimax-tdd-repo
     claude-fable-5: code-fable-tdd-repo

model-registry/minimax-m31-flash-max.yaml

@@ -0,0 +1,21 @@
+# agent-runtime-framework — model-registry/ (CRS tag: model-registry)
+# Max-effort Flash; replaces minimax-m3-max in planning/draft-heavy and explicit-model nodes.
+# PLACEHOLDER scores/costs: scores are copied from the M3 entry this replaces and
+# costs are set just below it so cheapest-wins (tiers.yaml) selects Flash over
+# the retained M3 entries. Revisit with observed scores and published pricing.
+name: minimax-m31-flash-max
+endpoint: minimax-m31-flash-max
+provider: minimax
+family: minimax
+scores:
+  complexity: 10
+  test_pass_rate: 9
+  creativity: 10
+  spec_adherence: 9
+  cost_efficiency: 5
+  context_utilisation: 10
+cost_per_1k_tokens:
+  input: 0.0025
+  output: 0.0125
+source: benchmark
+harness_version: null

model-registry/minimax-m31-flash-medium.yaml

@@ -0,0 +1,21 @@
+# agent-runtime-framework — model-registry/ (CRS tag: model-registry)
+# Medium-effort Flash; replaces minimax-m3 in the build/integrate/spec-test/coding tiers.
+# PLACEHOLDER scores/costs: scores are copied from the M3 entry this replaces and
+# costs are set just below it so cheapest-wins (tiers.yaml) selects Flash over
+# the retained M3 entries. Revisit with observed scores and published pricing.
+name: minimax-m31-flash-medium
+endpoint: minimax-m31-flash-medium
+provider: minimax
+family: minimax
+scores:
+  complexity: 8
+  test_pass_rate: 9
+  creativity: 8
+  spec_adherence: 9
+  cost_efficiency: 6
+  context_utilisation: 9
+cost_per_1k_tokens:
+  input: 0.00125
+  output: 0.00625
+source: benchmark
+harness_version: null

model-registry/minimax.yaml

@@ -1,15 +0,0 @@
-name: minimax
-endpoint: minimax
-provider: minimax
-scores:
-  complexity: 10
-  test_pass_rate: 9
-  creativity: 10
-  spec_adherence: 9
-  cost_efficiency: 3
-  context_utilisation: 10
-cost_per_1k_tokens:
-  input: 0.0
-  output: 0.0
-source: benchmark
-harness_version: null

model-registry/tiers.yaml

@@ -22,7 +22,7 @@ tiers:
       context_utilisation: 9
     qualifies_today:
       - deepseek-v4flash       # benchmark-inflated 9s; hold to allowed_models ceiling
-      - minimax
+      - minimax-m31-flash-max
       - claude-opus-5
     note: >
       WF-SCH-14 allowed_models ceiling remains authoritative until M7 observed-score
@@ -37,7 +37,8 @@ tiers:
       spec_adherence: 9
       test_pass_rate: 9
     qualifies_today:
-      - minimax
+      - minimax-m31-flash-max
+      - minimax-m31-flash-medium
       - claude-sonnet-5
       - claude-opus-5
 
@@ -51,7 +52,8 @@ tiers:
     qualifies_today:
       - airouter-qwen3
       - airouter-deepseekv4flash
-      - minimax
+      - minimax-m31-flash-max
+      - minimax-m31-flash-medium
       - claude-sonnet-5
       - claude-opus-5
       - claude-fable-5

models/minimax-m31-flash-max.yaml

@@ -0,0 +1,21 @@
+# agent-runtime-framework — models/ (CRS tag: models)
+# MiniMax-M3.1-Flash-Preview at reasoning_effort max. Always thinks (thinking
+# cannot be disabled; effort is the only depth control). The claude runner maps
+# default_flags.reasoning_effort to `--effort` (agent-runtimes ME-6a). Verified
+# 2026-10-13: the container MiniMax key reaches this model on both
+# api.minimax.io and api.minimaxi.chat; 1M context window.
+# Endpoint name MUST equal the registry name (MR-12).
+name: minimax-m31-flash-max
+provider: minimax
+family: minimax
+endpoint:
+  type: cloud_api
+  base_url: "https://api.minimax.io/anthropic"
+  api_key_env: MINIMAX_API_KEY
+runner: claude
+default_flags:
+  model: "MiniMax-M3.1-Flash-Preview"
+  reasoning_effort: max
+provider_context_ref: "minimax/v1"
+provider_secret_ref: "MINIMAX_API_KEY"
+session_capable: false

models/minimax-m31-flash-medium.yaml

@@ -0,0 +1,21 @@
+# agent-runtime-framework — models/ (CRS tag: models)
+# MiniMax-M3.1-Flash-Preview at reasoning_effort medium. Always thinks (thinking
+# cannot be disabled; effort is the only depth control). The claude runner maps
+# default_flags.reasoning_effort to `--effort` (agent-runtimes ME-6a). Verified
+# 2026-10-13: the container MiniMax key reaches this model on both
+# api.minimax.io and api.minimaxi.chat; 1M context window.
+# Endpoint name MUST equal the registry name (MR-12).
+name: minimax-m31-flash-medium
+provider: minimax
+family: minimax
+endpoint:
+  type: cloud_api
+  base_url: "https://api.minimax.io/anthropic"
+  api_key_env: MINIMAX_API_KEY
+runner: claude
+default_flags:
+  model: "MiniMax-M3.1-Flash-Preview"
+  reasoning_effort: medium
+provider_context_ref: "minimax/v1"
+provider_secret_ref: "MINIMAX_API_KEY"
+session_capable: false

task-templates/minimax-code-repo.yaml

@@ -1,6 +1,6 @@
 name: minimax-code-repo
 description: "MiniMax code agent — repo clone, agent-repo branch persistence"
-model: minimax
+model: minimax-m31-flash-max
 harness: code-minimax-repo/v1
 requires_tags: [claude-code, coding-agent, git-access, project-repo]
 required_params: [repo_url, agent_repo_url]

task-templates/minimax-code-tdd.yaml

@@ -1,6 +1,6 @@
 name: minimax-code-tdd
 description: "MiniMax code agent with TDD enforcement — tests are read-only and must pass before the task finishes"
-model: minimax
+model: minimax-m31-flash-max
 harness: code-minimax-tdd-repo/v1
 requires_tags: [coding-agent, git-access, project-repo, tdd-gate]
 required_params: [repo_url, agent_repo_url]

task-templates/minimax-planning.yaml

@@ -1,6 +1,6 @@
 name: minimax-planning
 description: "MiniMax planning agent — specs, reviews, architecture docs"
-model: minimax
+model: minimax-m31-flash-max
 harness: planning-minimax-repo/v1
 requires_tags: [claude-code, planning-agent, git-access, project-repo]
 required_params: [repo_url, agent_repo_url]

task-templates/minimax-security-review.yaml

@@ -1,6 +1,6 @@
 name: minimax-security-review
 description: "MiniMax security review agent — security audit via Anthropic-compatible proxy"
-model: minimax
+model: minimax-m31-flash-max
 harness: security-review-minimax-repo/v1
 requires_tags: [claude-code, security-reviewer, best-practices, git-access, project-repo]
 required_params: [repo_url, agent_repo_url]

task-templates/minimax-tdd-impl.yaml

@@ -1,6 +1,6 @@
 name: minimax-tdd-impl
 description: "MiniMax TDD implementation agent — tests locked read-only, cannot be modified by agent"
-model: minimax
+model: minimax-m31-flash-max
 harness: minimax-tdd-impl/v1
 requires_tags: [claude-code, coding-agent, git-access, project-repo]
 required_params: [repo_url, agent_repo_url]

task-templates/minimax-test-writer.yaml

@@ -1,6 +1,6 @@
 name: minimax-test-writer
 description: "MiniMax test writing agent — comprehensive tests with edge cases"
-model: minimax
+model: minimax-m31-flash-max
 harness: test-writing-minimax-repo/v1
 requires_tags: [claude-code, test-writer, best-practices, git-access, project-repo]
 required_params: [repo_url, agent_repo_url]

task-templates/workflow/bug-verify-sonnet@1.yaml

@@ -1,103 +0,0 @@
-name: bug-verify-sonnet
-version: 1
-description: Read-only verification of an ai-resolve-tagged bug using Sonnet planning
-runtime: planning-sonnet-repo
-input:
-  state: verify-pending
-  tags_required: [ai-resolve]
-  tags_forbidden: []
-  required_artifacts: []
-output:
-  state: 'verified|rejected'
-scope_budget:
-  files_modified_max: 1
-  lines_diff_max: 400
-  wall_clock_seconds_max: 1800
-nodes:
-  # NOTE (2026-09-09, work item 88ecd23e): the `verdict-tx` script node that
-  # used to follow this one has been REMOVED. `kind: script` nodes are
-  # harness-exempt (WF-COMP-6 / MN-38) — they run in their own empty
-  # container with no credentialed way to clone the agent's branch or read
-  # /workspace/verdict, so verdict-tx never actually worked (it always ran
-  # against its own empty /workspace). Replaced with the "agent
-  # self-transition" pattern (spec/typed-workflows.md): this node's own
-  # `cp-service-token/v1` layer (via the `planning-sonnet-repo` composite)
-  # gives it CP write credentials, so the agent PATCHes its own item's
-  # flow_state directly as its final step, instead of relying on a second
-  # node to read a hand-off file. finalize_run's output.state auto-advance
-  # is intentionally disabled for this template (its `output.state` is the
-  # pipe-separated "verified|rejected" — finalize.py treats any
-  # pipe-separated output.state as schema-validation-only, never a literal
-  # value to PATCH, so it never clobbers the value set below).
-  - id: agent
-    kind: agent
-    template: planning-sonnet-repo
-    model: sonnet
-    runtime_env:
-      AGENT_REPO_URL: "git@gitea.oreillyit.nz-ai-enablement:skynet/agent-runtimes-agents.git"
-      AGENT_UPSTREAM_REPO_URL: "git@gitea.oreillyit.nz-ai-enablement:skynet/agent-runtimes.git"
-      AGENT_UPSTREAM_BRANCH: "main"
-      AGENT_BRANCH: "bug-verify/${item.uuid}"
-      CP_URL: "https://agents.oreillyit.nz/api"
-    prompt: |
-      The agent-runtimes project repo is cloned at /workspace/project.
-      Change to that directory before doing any work.
-
-      Verify the following bug report. This is a READ-ONLY investigation —
-      do NOT modify any source code. Your only file changes are the
-      verification artifact described below.
-
-      **Title**: ${item.title}
-
-      **Bug report**:
-      ${item.body}
-
-      Steps:
-      1. cd /workspace/project
-      2. Investigate the codebase to determine whether the reported bug is
-         real, reproducible, and correctly diagnosed. Do not make code
-         changes — this is verification, not a fix.
-      3. Write your findings to
-         spec/bugs/<slug>-verification.md (derive <slug> from this item's
-         own slug) with exactly these sections, in order:
-         - `## Reproduction` — concrete steps or evidence confirming (or
-           failing to confirm) the reported symptom
-         - `## Root-Cause Hypothesis` — the mechanism you believe causes it,
-           with file/line references
-         - `## Verdict` — exactly one of `verified` or `rejected`
-         - `## Rationale` — why you reached that verdict
-      4. Commit and push the verification artifact on this branch.
-      5. Transition this work item's own flow_state to your verdict by
-         PATCHing the control plane directly — you have a scoped CP
-         write credential for exactly this (`CP_SERVICE_TOKEN_FILE`,
-         `CP_URL`). Set `flow_state` to exactly `verified` or exactly
-         `rejected` (matching the `## Verdict` you wrote above) — no other
-         value is accepted. Run:
-
-         ```
-         python3 - <<'PY'
-         import json, os, urllib.request, uuid
-
-         VERDICT = "verified"  # or "rejected" — must match your ## Verdict
-         ITEM_UUID = "${item.uuid}"
-
-         token = open(os.environ["CP_SERVICE_TOKEN_FILE"]).read().strip()
-         url = os.environ["CP_URL"].rstrip("/") + f"/v1/work-items/{ITEM_UUID}"
-         body = json.dumps({
-             "flow_state": VERDICT,
-             "description": f"bug-verify-sonnet@1: verdict={VERDICT}",
-         }).encode()
-         req = urllib.request.Request(url, data=body, method="PATCH", headers={
-             "Authorization": f"Bearer {token}",
-             "Content-Type": "application/json",
-             "Idempotency-Key": str(uuid.uuid4()),
-         })
-         with urllib.request.urlopen(req, timeout=30) as resp:
-             print(resp.status, resp.read().decode())
-         PY
-         ```
-
-         Replace `VERDICT` with your actual verdict before running. This is
-         the ONLY code you may run outside the investigation itself — do
-         not use this credential for anything else.
-      6. Call task_complete only after the PATCH above returns 200.

task-templates/workflow/frozen-fixture-emit@1.yaml

@@ -24,7 +24,6 @@ nodes:
     kind: script
     cmd: scripts/workflow/state-tx
     cp_write: true
-    on_fail: continue
     runtime_env:
       CP_URL: "https://agents.oreillyit.nz/api"
       STATE_TX_NEW_STATE: test-skeleton-ready

task-templates/workflow/plan-review-loop@1.yaml

@@ -46,7 +46,7 @@ nodes:
   - id: review-minimax
     kind: agent
     role: plan-review
-    model: minimax-m2
+    model: minimax-m31-flash-max
     runtime_env:
       AGENT_REPO_URL: "git@gitea.oreillyit.nz-ai-enablement:skynet/agent-runtimes-agents.git"
       AGENT_BRANCH: "task/${item.uuid}"

task-templates/workflow/scope-decompose-sonnet@1.yaml

@@ -1,68 +0,0 @@
-name: scope-decompose-sonnet
-version: 1
-description: Scope and decompose a spec into tagged tasks using Sonnet planning
-runtime: spec-writing-sonnet-repo
-input:
-  state: spec-reviewed-arch
-  tags_forbidden: []
-  required_artifacts: []
-output:
-  state: scoped-tagged
-scope_budget:
-  files_modified_max: 10
-  lines_diff_max: 500
-  wall_clock_seconds_max: 2400
-nodes:
-  - id: agent
-    kind: agent
-    # Decomposition is spec-test-tier work (claude/stage-routing.md): Sonnet,
-    # not airouter. 2026-07-04 E2E: Qwen3.6 empty-delivered this stage (task
-    # "succeeded" touching only .gitignore, zero work items created) — the
-    # template NAME never set the model; the node model: field does.
-    template: spec-writing-sonnet-repo
-    # cli defaults to "claude" — the anthropic-cloud OAuth composite drives
-    # the claude CLI; "cli: agentic" is the OpenAI-compatible runner and
-    # fails fast with "OPENAI_BASE_URL required" (E2E 2026-07-04).
-    model: sonnet
-    runtime_env:
-      AGENT_REPO_URL: "git@gitea.oreillyit.nz-ai-enablement:skynet/agent-runtimes-agents.git"
-      AGENT_UPSTREAM_REPO_URL: "git@gitea.oreillyit.nz-ai-enablement:skynet/agent-runtimes.git"
-      AGENT_UPSTREAM_BRANCH: "main"
-      AGENT_BRANCH: "spec/${item.uuid}"
-      CP_URL: "https://agents.oreillyit.nz/api"
-      # AR-38: this stage's deliverable is work items POSTed to the CP, not a
-      # repo diff — the empty-deliverable audit would fail every success.
-      AGENT_EMPTY_DELIVERABLE_CHECK: "false"
-    prompt: |
-      The agent-runtimes project repo is cloned at /workspace/project.
-      Change to that directory before doing any work.
-
-      Find the spec file in /workspace/project/spec/ (the spec was
-      written in an earlier pipeline stage — look for a file whose name matches
-      the concept title '${item.title}' as a slug, e.g. spec/<slug>.md).
-      Read that spec carefully.
-
-      Write a decomposition plan to /workspace/.agent-output/decompose.json with
-      exactly this structure:
-      {
-        "project_id": "${item.project_id}",
-        "space_id": "${item.space_id}",
-        "parent_item_uuid": "${item.uuid}",
-        "tasks": [
-          {
-            "title": "...",
-            "description": "...",
-            "flow_state": "needs-tests",
-            "spec_ids": ["WT-EVAL-1", "..."],
-            "test_files": ["tests/triggers/test_schema.py", "..."],
-            "tags": ["airouter-eligible"]
-          }
-        ]
-      }
-
-      Include "airouter-eligible" in tags only for single-file, self-contained
-      tasks that Qwen3.6 can handle alone. Omit tags for complex multi-file tasks.
-
-      After writing decompose.json, run this command to create the work items:
-
-      python3 -m lib.cp_cli planner items decompose ${item.uuid} --file /workspace/.agent-output/decompose.json

task-templates/workflow/spec-review-loop@1.yaml

@@ -54,7 +54,7 @@ nodes:
   - id: review-minimax
     kind: agent
     role: spec-review
-    model: minimax-m2
+    model: minimax-m31-flash-max
     runtime_env:
       AGENT_REPO_URL: "git@gitea.oreillyit.nz-ai-enablement:skynet/agent-runtimes-agents.git"
       AGENT_BRANCH: "spec/${item.uuid}"

task-templates/workflow/test-validate@1.yaml

@@ -29,12 +29,10 @@ nodes:
     kind: script
     cmd: scripts/workflow/preflight-test-fail-mode
     cp_write: false
-    on_fail: continue
   - id: state-tx
     kind: script
     cmd: scripts/workflow/state-tx
     cp_write: true
-    on_fail: continue
     runtime_env:
       CP_URL: "https://agents.oreillyit.nz/api"
       STATE_TX_NEW_STATE: test-validated

task-templates/workflow/test-write-review-loop@1.yaml

@@ -21,7 +21,7 @@ nodes:
     role: test-write-review  # M36 v0.6 (bug 69370eda fix): role + model replaces
                              # the ungrounded `template: minimax-test-writer`
                              # (a composite that never existed in the catalog).
-    model: minimax-m2
+    model: minimax-m31-flash-max
     runtime_env:
       # WF-COMP-15: stage agents share one per-item work branch so their work
       # accumulates across iterations.

tests/invariants/test_cp_docs_token_layer.py

@@ -0,0 +1,56 @@
+"""cp-docs-token/v1 context: env-only step-docs layer (PLN-CLI-16, D-40, AU-59).
+
+The per-step wfcd_ token and CA file are delivered by the dispatcher. The layer
+sets the env contract and ships one context doc; no ESO mount, no init script,
+and no reference to the cp-service-token layer.
+"""
+from __future__ import annotations
+
+from pathlib import Path
+
+import yaml
+
+_LAYER = (
+    Path(__file__).resolve().parents[2] / "harnesses" / "contexts" / "cp-docs-token" / "v1"
+)
+
+
+def _doc() -> dict:
+    return yaml.safe_load((_LAYER / "harness.yaml").read_text())
+
+
+def test_cp_docs_token_layer_parses_with_env_keys():
+    doc = _doc()
+    assert doc["kind"] == "context"
+    assert doc["name"] == "cp-docs-token"
+    assert doc["version"] == 1
+    assert doc["requires"] == []
+    assert doc["provides"] == ["cp-docs-token"]
+    assert doc["env"] == {
+        "CP_DOCS_TOKEN_FILE": "/run/agent/cp-docs/token",
+        "CP_DOCS_CA_FILE": "/run/agent/cp-docs/ca.crt",
+    }
+
+
+def test_cp_docs_token_layer_declares_no_secrets_and_no_init():
+    doc = _doc()
+    assert not doc.get("secrets_required")
+    assert not doc.get("scripts")
+    assert not (_LAYER / "init.sh").exists()
+
+
+def test_cp_docs_token_layer_context_file_mapping():
+    doc = _doc()
+    assert doc["context_files"] == [
+        {
+            "source": "./CLAUDE.md",
+            "target": "/opt/harness/context/cp-docs-token/CLAUDE.md",
+        }
+    ]
+    assert (_LAYER / "CLAUDE.md").is_file()
+
+
+def test_cp_docs_token_layer_never_mentions_cp_service_token():
+    for path in _LAYER.rglob("*"):
+        if path.is_file():
+            assert "cp-service-token" not in path.read_text(), path

tests/invariants/test_no_legacy_script_composite.py

@@ -0,0 +1,32 @@
+"""Track R / R3b: the legacy AU-51 bearer composite and layer stay deleted.
+
+``workflow-script/v1`` (the legacy script composite) and the
+``cp-service-token/v1`` context were removed once the control plane stopped
+falling back to them (agent-runtimes 0.0.5.1890). Nothing may re-introduce
+either, and no live composite may layer the bearer context.
+"""
+
+from pathlib import Path
+
+import yaml
+
+_ROOT = Path(__file__).resolve().parents[2]
+_HARNESSES = _ROOT / "harnesses"
+
+
+def test_legacy_script_composite_directory_is_gone():
+    assert not (_HARNESSES / "composites" / "workflow-script").exists()
+
+
+def test_cp_service_token_context_directory_is_gone():
+    assert not (_HARNESSES / "contexts" / "cp-service-token").exists()
+
+
+def test_no_composite_layers_cp_service_token():
+    offenders = []
+    for path in sorted((_HARNESSES / "composites").glob("*/v*/harness.yaml")):
+        doc = yaml.safe_load(path.read_text()) or {}
+        for layer in doc.get("layers") or []:
+            if str(layer.get("context", "")).startswith("cp-service-token"):
+                offenders.append(str(path.relative_to(_ROOT)))
+    assert offenders == []

tests/invariants/test_state_tx_runtime_env.py

@@ -93,9 +93,7 @@ def test_cp_url_templates_discovered():
     assert {
         "test-validate@1.yaml",
         "frozen-fixture-emit@1.yaml",
-        "scope-decompose-sonnet@1.yaml",
         "scope-decompose-sonnet@2.yaml",
-        "bug-verify-sonnet@1.yaml",
         "bug-verify-sonnet@2.yaml",
     } <= names
 

tests/invariants/test_workflow_script_composite_layers.py

@@ -1,35 +0,0 @@
-"""workflow-script/v1 composite: exact layer order and credential env export.
-
-Script nodes get their CP write credential only through this composite
-(the legacy WF-COMP-26 mount is removed), so the layering is load-bearing.
-"""
-from __future__ import annotations
-
-from pathlib import Path
-
-import yaml
-
-_ROOT = Path(__file__).resolve().parents[2] / "harnesses"
-
-
-def _load(kind_dir: str, name: str) -> dict:
-    return yaml.safe_load((_ROOT / kind_dir / name / "v1" / "harness.yaml").read_text())
-
-
-def test_workflow_script_layers_shared_variables_cp_service_token_repo_readonly():
-    doc = _load("composites", "workflow-script")
-    assert doc["layers"] == [
-        {"context": "shared-variables/v1"},
-        {"context": "cp-service-token/v1"},
-        {"context": "repo-readonly/v1"},
-    ]
-
-
-def test_workflow_script_env_exports_cp_service_token_file():
-    doc = _load("composites", "workflow-script")
-    env: dict = dict(doc.get("env") or {})
-    for layer in doc["layers"]:
-        (kind, ref), = layer.items()
-        name = ref.split("/")[0]
-        env.update(_load(kind + "s", name).get("env") or {})
-    assert env["CP_SERVICE_TOKEN_FILE"] == "/run/agent/cp-service-token/token"

tests/invariants/test_workflow_script_composite_resolves.py

@@ -1,227 +0,0 @@
-"""Workflow-script harness composition invariant (spec/typed-workflows.md WF-COMP-31
-+ spec/workflow-variables.md WV-19 + spec/harness.md HD-14).
-
-The two new framework layers — ``harnesses/contexts/shared-variables/v1/`` and
-``harnesses/composites/workflow-script/v1/`` — must exist on the tree, must parse,
-and the composite must resolve every layer it references to a real file on disk.
-This test guards against the most common regression: a missing or renamed layer
-that would only surface at the dispatcher's ``compose_harness`` call, where it
-costs an entire task's worth of wasted dispatch round-trip.
-
-Spec: /home/paul/dev/claude/projects/agent-runtimes/spec/workflow-variables.md
-      (WV-19, WV-20, WF-COMP-31), spec/harness.md (HD-14).
-
-Not committed by the authoring agent per its instructions — written here so it
-can run locally against the current tree.
-"""
-from __future__ import annotations
-
-from pathlib import Path
-
-import pytest
-import yaml
-
-_HARNESSES_ROOT = Path(__file__).resolve().parents[2] / "harnesses"
-
-# Composite layer entries use the singular kind as the dict key. The kind maps
-# to a top-level directory by adding "s".
-_KIND_TO_DIR = {
-    "context": "contexts",
-    "capability": "capabilities",
-    "composite": "composites",
-}
-
-
-def _load_layer(kind_dir: str, name: str) -> dict:
-    p = _HARNESSES_ROOT / kind_dir / name / "v1" / "harness.yaml"
-    return yaml.safe_load(p.read_text(encoding="utf-8"))
-
-
-def test_shared_variables_layer_exists_and_parses():
-    """``harnesses/contexts/shared-variables/v1/harness.yaml`` must exist,
-    parse as YAML, and carry ``provides: [shared-variables]`` (the HD-14 tag).
-
-    Spec: spec/workflow-variables.md WV-19; spec/harness.md HD-14.
-    """
-    parsed = _load_layer("contexts", "shared-variables")
-    assert parsed["kind"] == "context"
-    assert parsed["name"] == "shared-variables"
-    assert parsed["version"] == 1
-    assert parsed["provides"] == ["shared-variables"], (
-        "WV-19: provides MUST contain the 'shared-variables' tag — HD-14's "
-        "extra_layers dedup key, and the tag the dispatcher's payload-marker "
-        "injection matches against."
-    )
-    # WV-20: this layer's OWN env: declaration sets the two WORKFLOW_VARIABLES_*
-    # vars (NOT the entrypoint, NOT the container backend).
-    env = parsed["env"]
-    assert env["WORKFLOW_VARIABLES_ROOT"] == "/variables"
-    assert "WORKFLOW_VARIABLES_SELF" in env, (
-        "WV-20: WORKFLOW_VARIABLES_SELF must be set by the layer's env: "
-        "(templated per-node at harness-composition time on the dispatcher)"
-    )
-    # Agent-facing context: a CLAUDE.md must be mounted under /opt/harness/context/.
-    files = parsed.get("context_files", [])
-    targets = [f["target"] for f in files]
-    assert any(t.startswith("/opt/harness/context/shared-variables/") for t in targets), (
-        "WV-19: agent-facing context must be mounted under /opt/harness/context/"
-        "(the HC-7 namespacing point) so the agent sees it at startup."
-    )
-
-
-def test_shared_variables_claude_md_present_and_substantial():
-    """The agent-facing CLAUDE.md must be a non-trivial markdown document
-    (HC-7 says it is read by the agent, not a placeholder).
-
-    Spec: spec/workflow-variables.md WV-19; spec/harness.md HC-7.
-    """
-    md = _HARNESSES_ROOT / "contexts" / "shared-variables" / "v1" / "CLAUDE.md"
-    assert md.exists(), "WV-19: shared-variables/v1 must ship a CLAUDE.md"
-    text = md.read_text(encoding="utf-8")
-    assert text.count("\n") >= 30, (
-        f"WV-19: the CLAUDE.md must explain the /variables/ contract to the "
-        f"agent — got {text.count(chr(10))} lines, expected >=30."
-    )
-    # Spot-check the four high-value pieces of content the agent MUST see:
-    assert "/variables/" in text, "must explain the filesystem root"
-    assert "WORKFLOW_VARIABLES_SELF" in text, "must document the env vars"
-    assert "wfvar" in text, "must point at the CLI helper (WV-21)"
-    assert "reserved" in text.lower(), "must list the reserved-name families (WV-6)"
-
-
-def test_workflow_script_composite_resolves():
-    """``harnesses/composites/workflow-script/v1/harness.yaml`` must declare
-    exactly the two layers the spec names (WF-COMP-31) and every referenced
-    file must exist on disk.
-
-    Spec: spec/typed-workflows.md WF-COMP-31 ("Internal composite layered from
-    ``shared-variables/v1`` + ``cp-service-token/v1``").
-    """
-    parsed = _load_layer("composites", "workflow-script")
-    assert parsed["kind"] == "composite"
-    assert parsed["name"] == "workflow-script"
-    assert parsed["version"] == 1
-    assert isinstance(parsed["layers"], list) and len(parsed["layers"]) >= 2, (
-        "WF-COMP-31: the composite must layer at least shared-variables/v1 + "
-        "cp-service-token/v1; missing one breaks the credential env seam."
-    )
-
-    refs = []
-    for layer in parsed["layers"]:
-        assert isinstance(layer, dict) and len(layer) == 1, (
-            f"each layer entry must be a single-key dict (kind: ref), got {layer}"
-        )
-        kind_name, ref = next(iter(layer.items()))
-        assert kind_name in _KIND_TO_DIR, f"unknown layer kind {kind_name!r}"
-        assert "/" in ref, f"layer ref must be '<name>/v<N>', got {ref!r}"
-        name, ver = ref.split("/")
-        assert ver.startswith("v") and ver[1:].isdigit(), (
-            f"layer ref version must match v<N>, got {ver!r}"
-        )
-        kind_dir = _KIND_TO_DIR[kind_name]
-        layer_path = _HARNESSES_ROOT / kind_dir / name / ver / "harness.yaml"
-        assert layer_path.exists(), (
-            f"WF-COMP-31: composite references a layer that does not exist on "
-            f"the tree: {layer_path.relative_to(_HARNESSES_ROOT)}"
-        )
-        layer_data = yaml.safe_load(layer_path.read_text(encoding="utf-8"))
-        assert layer_data["name"] == name, (
-            f"name mismatch in {layer_path}: file says {layer_data['name']!r}, "
-            f"ref says {name!r}"
-        )
-        refs.append((kind_name, name, ver, layer_data))
-
-    # WF-COMP-31: the two SPECIFIED layers must both be present.
-    ref_names = {(kind, name) for kind, name, _, _ in refs}
-    assert ("context", "shared-variables") in ref_names, (
-        "WF-COMP-31: workflow-script/v1 MUST layer shared-variables/v1 "
-        "(WV-19 — the agent-facing /variables/ contract)"
-    )
-    assert ("context", "cp-service-token") in ref_names, (
-        "WF-COMP-31 / AU-51 credential env seam: workflow-script/v1 MUST "
-        "layer cp-service-token/v1 so CP_SERVICE_TOKEN_FILE is in the "
-        "entrypoint process env before shell_runner runs."
-    )
-
-
-def test_workflow_script_composite_brings_no_model_or_endpoint_layer():
-    """``kind: script`` nodes run via the shell runner — the composite MUST
-    NOT bring agent-only layers (no model, endpoint, CLAUDE.md context that
-    drives the agent runner, etc.). This is a structural regression gate: a
-    future refactor that adds e.g. ``anthropic-cloud-paul-oauth/v1`` here
-    would silently widen the script-node attack surface.
-
-    Spec: spec/typed-workflows.md WF-COMP-31 ("the flag-on kind: script
-    harness ... MUST NOT bring agent-only layers").
-    """
-    parsed = _load_layer("composites", "workflow-script")
-    forbidden_substrings = (
-        # cloud-auth contexts the agent runner needs
-        "anthropic-cloud", "minimax", "qwen-code", "deepseek-code",
-        "alibaba-model-studio", "fable-low-effort", "z-ai",
-        # the model-bearing capability / context layers
-        "code-methodology", "spec-writing", "planning", "integration",
-        # harness composites that bundle a model
-        "code-", "planning-", "spec-writing-", "best-practices-",
-        "sonnet-", "opus-", "minimax-", "haiku-", "airouter-", "fable-",
-    )
-    layer_names = [
-        next(iter(layer.values())).split("/")[0]
-        for layer in parsed["layers"]
-    ]
-    bad = [
-        n for n in layer_names
-        if any(sub in n for sub in forbidden_substrings)
-    ]
-    assert not bad, (
-        f"WF-COMP-31: workflow-script/v1 MUST NOT bring agent-only layers; "
-        f"found: {bad}"
-    )
-
-
-def test_cp_service_token_layer_already_sets_cp_service_token_file_env():
-    """WF-COMP-31 'AU-51 credential env seam': cp-service-token/v1 must set
-    ``CP_SERVICE_TOKEN_FILE`` in its env: declaration (the layer's init.sh
-    restages the ESO-mounted secret to an agent-readable 0600 copy at the
-    path CP_SERVICE_TOKEN_FILE points at). Verified here at the YAML level
-    rather than at runtime — the spec's claim is structural.
-
-    Spec: spec/typed-workflows.md WF-COMP-31; spec/harness.md AU-51/AU-51b
-    (bug b3a96acc).
-    """
-    parsed = _load_layer("contexts", "cp-service-token")
-    env = parsed["env"]
-    assert "CP_SERVICE_TOKEN_FILE" in env, (
-        "WF-COMP-31 'AU-51 credential env seam': cp-service-token/v1 must "
-        "declare CP_SERVICE_TOKEN_FILE in its env: block; without it the "
-        "shell_runner subprocess does not inherit the AU-51 credential."
-    )
-    fpath = env["CP_SERVICE_TOKEN_FILE"]
-    assert fpath.endswith("/token"), (
-        f"CP_SERVICE_TOKEN_FILE must point at the restaged 0600 file, not "
-        f"the raw 0400 ESO mount — got {fpath!r}"
-    )
-
-
-def test_workflow_script_composite_forge_credentials_are_read_only():
-    """Any layer in workflow-script/v1 that carries a forge credential MUST
-    declare ``write_targets: []`` (HC-11): a script node may read the project
-    repo but never push. The write-capable gitea-ssh/v1 layer must not be
-    present (bug 45a48685).
-
-    Spec: spec/harness.md HC-11, spec/typed-workflows.md WF-COMP-34.
-    """
-    parsed = _load_layer("composites", "workflow-script")
-    from tests.invariants.forge_credentials import _layer_has_forge_credentials
-
-    for layer in parsed["layers"]:
-        kind_name, ref = next(iter(layer.items()))
-        name = ref.split("/")[0]
-        assert name != "gitea-ssh", "workflow-script/v1 must not layer the write-capable gitea-ssh/v1 (bug 45a48685)"
-        kind_dir = _KIND_TO_DIR[kind_name]
-        sub = _load_layer(kind_dir, name)
-        if _layer_has_forge_credentials(sub):
-            assert sub.get("write_targets") == [], (
-                f"workflow-script/v1's layer {name} carries a forge credential "
-                f"but does not declare write_targets: [] (HC-11 read-only)."
-            )