feat(airouter): DeepSeek V4 Flash TDD composite + max-effort template

New code-airouter-deepseekv4flash-tdd-repo/v1 composite (deepseek
methodology + the same tdd-protect/tdd-gate/python-dev/tdd-file-lock
layers as the Qwen TDD composite), plus a task-template that sets
reasoning_effort=max via the new extra_body passthrough (agent-runtimes
TT-38). Model/endpoint/model-registry entries already existed; only
the TDD-capable harness and the template were missing.
This commit is contained in:
Paul O'Reilly
2026-07-03 12:36:43 +12:00
parent dab3db3727
commit 71f9cc0338
2 changed files with 50 additions and 0 deletions

View File

@@ -0,0 +1,31 @@
name: airouter-code-tdd-deepseekv4flash-max
description: "Airouter DeepSeek V4 Flash code agent (reasoning_effort=max) with TDD enforcement — agentic runner, tests read-only, must pass before finish"
harness: code-airouter-deepseekv4flash-tdd-repo/v1
model: DeepSeek-V4-Flash
runtime_cli: agentic
extra_body:
reasoning_effort: max
requires_tags: [agentic-runner, coding-agent, git-access, project-repo, best-practices, tdd-gate]
required_params: [repo_url, agent_repo_url]
labels: [airouter]
defaults:
timeout: 3600
max_turns: 100
pre_actions:
- type: clone
repo: "{{ repo_url }}"
branch: main
depth: 1
blocked_combinations:
- field: on_success
contains_type: commit_pr
message: "Use agent_repo instead of commit_pr — avoids F77 SSH alias URL bug"
warnings:
- "Airouter agents have no web search — all context must be in the repo or prompt"
- "Hard cap: 3 concurrent airouter tasks (DISPATCHER_MAX_TASKS=3 on airouter dispatcher)"
- "Test files are locked read-only at the OS level — agent cannot modify them"
- "Task will be marked failed if the test suite is not green on completion"
- "Use narrow, single-file or single-requirement tasks for best results"
- "allowed_failing_tests support requires M19 (pre-test hook) to be deployed"
- "reasoning_effort=max is forwarded via extra_body (TT-38, agent-runtimes commit pending) — requires airouter/DeepSeek to honor OpenAI-compatible extra_body passthrough; unverified against airouter.ch's specific proxy behavior as of 2026-07-03"
- "New model (released 2026-07): no track record in this codebase yet — treat as experimental until a few tasks land cleanly"