Files
custom-claude-skills/MEMORY.md
Paul O'Reilly e8e572e0af Add decompose + orchestrate skills with operational improvements
New skills for container agent task orchestration:
- /decompose: break tasks into subtasks with dependency graph, write .agent-tasks.json
- /orchestrate: check task state, launch container agents in git worktrees, poll for completion

Updated with learnings from first real run (agent-runtimes M3, 9 tasks):
- Prompts must end with "run pytest and fix failures"
- State import conventions explicitly in prompts
- Warn about uncommitted files before decomposing (worktrees need committed content)
- Copy dependency outputs into downstream worktrees before launching

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-03-24 23:29:43 +13:00

1.8 KiB

Custom Claude Skills — Memory

Skill Development Learnings

  • Skills in ~/.claude/skills/ are available across all projects (personal scope)
  • !command`` injection runs at skill load time — great for pre-fetching git logs, file contents, etc.
  • Inline skills (no context: fork) retain full conversation history — essential for reflection-type skills
  • allowed-tools reduces permission prompts by declaring what the skill needs upfront
  • Bang-command exit codes matter: !command`` fails the skill if the command exits non-zero, even if the output is valid. Scripts that use exit 1 as "issues found" (not error) need || true appended in the skill invocation.
  • Every binary in bang-commands needs allowed-tools coverage: ls, head, cat, hostname — each needs its own Bash(<binary> *) pattern or the skill fails to load
  • Broad Bash(git *) is acceptable when justified: orchestrate skill needs worktree/branch/merge/checkout — listing 6+ specific subcommands is worse than the broad pattern with a validator warning
  • Skills that launch containers need careful entrypoint handling: Use --entrypoint uid-wrapper.sh when running claude --print in agent containers — the default entrypoint expects AGENT_PAYLOAD

Task Orchestration Skills (2026-03-24)

Added /decompose and /orchestrate for parallel container agent work:

  • /decompose breaks tasks into subtasks with dependency graph → .agent-tasks.json
  • /orchestrate dispatches container agents in git worktrees, checks status, updates state
  • Designed for /loop 2m /orchestrate auto-polling pattern
  • Git worktrees provide isolation: each agent gets its own branch, no file conflicts
  • No hardcoded concurrency limit — user approves max_concurrent during decompose
  • Memory files created in claude-foundations/memory/skill-decompose.md and skill-orchestrate.md