Project Knowledge

SkillAI & models

Maintain a project's canonical terms and settled decisions that future work should reuse, and record follow-ups a session discovers but does not resolve. Use when project-specific terms are being clarified, when a choice that may constrain future work is being considered or settles, including during planning, or when a session applies a temporary workaround whose root cause stays open or observes an out-of-scope defect with evidence. Do not use for simple definitions, routine implementation details, or carrying out an already-settled decision.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Project Knowledge skill

What this skill tells your AI

The instructions your AI receives, as published by toy-crane/claude-code-playbook-template in .agents/skills/project-knowledge/SKILL.md and read by ahel’s review.

Resolve unclear project terms and preserve settled decisions so future work uses the same language and does not re-litigate the same trade-offs.

Resolve terms

Read GLOSSARY.md when it exists.

  • Resolve vague, overloaded, or conflicting project terms with the user.
  • Stress-test unclear relationships between domain concepts with concrete edge-case scenarios.
  • When a term is resolved, create GLOSSARY.md if needed and update it immediately using the glossary template.
  • Keep only project terms and their current definitions in GLOSSARY.md.

Preserve decisions

When docs/decisions/README.md exists, read it and load only the subject files relevant to the current work.

Record a project decision only when it is all of the following:

  1. Settled: the choice is no longer a proposal. The user confirmed the outcome, or it was chosen under authority the user explicitly delegated for this class of decision.
  2. Reusable: future work is likely to face the same question and should reuse this answer unless its reconsideration conditions are met.
  3. Non-obvious: without its rationale, future work could reasonably reopen the question or choose differently.
  4. A real trade-off: plausible alternatives existed, and preserving why they were rejected prevents the same evaluation from recurring.

Require evidence that a choice was intentional before treating it as settled. Implementation or lack of objection alone is insufficient.

For each qualifying decision, use the decision contract template to create or update its single subject file in docs/decisions/. Preserve only the context future work needs to apply the decision without repeating the original analysis.

Record follow-ups

Record a follow-up when a workaround applied in this session is temporary and its root cause stays open, or when an out-of-scope defect or suspected cause is observed with evidence. Work that closed inside the session, a guess without evidence, and a defect already fixed in the current change do not qualify.

Write each qualifying item at the moment of discovery to its own docs/follow-ups/<slug>.md using the follow-up template, naming the slug for the symptom. An item must let a later session act without this conversation. Reporting the discovery in conversation does not preserve it.

A follow-up records an open question, not a settled decision, so it does not enter GLOSSARY.md or a decision contract until its outcome settles on its own terms. Delete the file when the work ships or the item is promoted into a spec folder.

Protect project truth

  • If code conflicts with the user's statement, GLOSSARY.md, or a relevant decision contract, surface the mismatch instead of choosing silently. Code shows current behavior, but not whether that behavior was intentional.
  • A request to align documentation with code does not by itself confirm the decision the code implies.
  • If relevant sources disagree about a settled decision, leave project knowledge unchanged until the intent is explicitly clarified.
  • Treat a term or decision as preserved only after its target file is updated.

Signals

GitHub stars
60
Forks
8
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
project-knowledge-toy-crane
Source
github.com/toy-crane/claude-code-playbook-template