Cmon Compound

Do not write a solution doc for every trivial edit.

This skill is for reusable learning, not for logging every completed task.

Output Targets

Default target:

• docs/solutions/

Other valid targets when appropriate:

• docs/architecture/ for stable repo-level design decisions
• existing docs that need to be updated rather than duplicated

Default templates:

• templates/solutions/bug-solution-template.md for solved bug or failure learnings
• templates/solutions/pattern-solution-template.md for reusable practices, patterns, and guardrails

Use templates/compound/compound-context-template.md as the default shared context shape for multi-role compounding.

Workflow

Stage 1: Run trigger check

Start from:

• templates/compound/compound-run-manifest-template.md when operator control is needed
• templates/compound/trigger-checklist-template.md to classify the signal as mandatory, recommended, or skip

Stage 2: Build shared compound context

Assemble:

• completed work summary
• source artifacts
• trigger signals
• related existing solution docs

Stage 3: Dispatch lens curators

Dispatch in parallel:

• agents/compound/product-curator.md
• agents/compound/engineer-curator.md
• agents/compound/ops-curator.md

Use templates/compound/lens-invocation-template.md as the default wrapper for each curator call.

Stage 4: Synthesize the compound decision

Pass the shared context and curator outputs into:

• agents/compound/compound-synthesizer.md

Use templates/compound/synthesizer-input-template.md as the default synthesis structure.

Stage 5: Write or update durable knowledge

Decide whether to:

• create a new bug doc
• create a new pattern doc
• update an existing doc
• skip if the signal is too weak

If the pass discovers multiple stale, overlapping, or contradictory docs that go beyond a narrow in-place update:

• recommend a targeted cmon:refresh-knowledge follow-up instead of widening the current compounding pass into a broad maintenance sweep

Preferred Categories of Learning

• solved bug with reusable root cause
• implementation pattern worth reusing
• review rule discovered through failure
• operational constraint or rollout gotcha
• planning heuristic that should recur

Hard Rules

• Do not ignore a mandatory trigger once it is established
• Do not create duplicates when an existing doc should be updated
• Do not write vague "we learned a lot" summaries
• Do not capture chat-only detail without explaining the reusable lesson
• Do not turn trivial work into noisy solution docs
• Do not collapse product, engineering, and operations learning into one unlabeled generic lesson when the multi-lens pipeline is available

Handoff

Record the handoff using:

• templates/workflow/stage-transition-decision-template.md

Typical transition decisions:

• proceed -> complete

  • after compounding when the durable knowledge state is updated

• revise -> cmon:compound

  • if the learning artifact is real but still needs tightening

• block

  • if compounding cannot finish because the source learning is still unresolved or contradictory

Future runs of cmon:understand, cmon:think, cmon:plan, cmon:challenge, and cmon:verify should be able to discover the artifact.