From agent-skills
Reconstructs the ordered workflow behind a shipped artifact from ax session evidence, commits, and PR context. Use to understand what made a feature, fix, or PR work.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-skills:ax-extract-workflowThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Shipping something once is not the same as knowing how to repeat it. This skill reconstructs the ordered workflow behind a past artifact: the framing decisions, implementation path, review gates, verification evidence, and user steering points that made it land.
Shipping something once is not the same as knowing how to repeat it. This skill reconstructs the ordered workflow behind a past artifact: the framing decisions, implementation path, review gates, verification evidence, and user steering points that made it land.
Use ax as the primary evidence source when it is installed, because it can connect sessions, turns, skills, tool calls, commits, and cost data. If ax is unavailable, use git history, PR comments, issue threads, CI logs, and local notes, but label the reconstruction as lower confidence.
NOT for:
debugging-and-error-recovery firstspec-driven-development and planning-and-task-breakdownIdentify the artifact the user wants to reconstruct:
| Anchor | Evidence to gather |
|---|---|
| Commit SHA | ax sessions near <sha> plus git show <sha> |
| PR URL or number | PR description, comments, review threads, checks, and commits |
| Date or release | ax sessions around <date> plus release notes or tags |
| Topic or feature name | ax recall "<topic>" to find candidate sessions and commits |
If the anchor is ambiguous, ask the user to choose before narrating.
Collect the smallest window that explains the artifact:
When using ax, prefer:
ax sessions near <sha> --json
ax sessions show <id> --json
ax sessions show <id> --by-role
ax recall "<topic>" --sources=turn,commit,skill --scope=here
Turn the evidence into an ordered arc:
Each step must cite evidence: session id, commit SHA, PR comment, test output, or file path.
Pull out 2-5 decisions that changed the outcome. Good decisions are specific:
Avoid fake lessons like "communicate clearly" or "write better tests." If the lesson could apply to any project, it is not extracted enough.
Return three sections:
Keep it factual. Do not turn uncertainty into a confident story.
| Rationalization | Why It Is Wrong |
|---|---|
| "The final commit tells the story." | The commit omits failed paths, user steering, verification churn, and subagent work. |
| "I can infer the process from the diff." | Diffs show what changed, not why the sequence worked. Use session, PR, and test evidence. |
| "Nearby sessions are probably relevant." | Time proximity is weak evidence. Include only sessions tied to the artifact by files, topic, commit, or explicit mention. |
| "A polished narrative is enough." | The output must be auditable. Every important claim needs a source. |
Before finishing, confirm:
npx claudepluginhub ishandutta2007/awesome-agent-skillsReconstructs the workflow behind a coding-agent artifact by analyzing local ax sessions, commits, skills, and tool traces. Useful when asked how something was built.
Reconstructs the ordered skill workflow behind a shipped artifact from past sessions, using ax and commit or date anchors.
Orchestrates Entire skills to read historical checkpoints for understanding code evolution, provenance, and intent. Delegate to sub-skills for explain, search, review, replay, and more.