ax-extract-workflow

Imported from yangsonhung/awesome-agent-skills (skills/en/ax-extract-workflow/SKILL.md). Install upstream with npx skills add yangsonhung/awesome-agent-skills --skill ax-extract-workflow. Copyright stays with the author.

ax Extract Workflow

Overview

Reconstruct the workflow behind a shipped artifact from local ax data. Resolve an anchor, find the relevant sessions, inspect turns, tools, skills, commits, and delegations, then write a compact account of what happened.

This skill assumes ax is installed, session data has been ingested, and the local ax database is reachable.

When to Use

Use this skill when the user asks for:

• How a feature, PR, demo, or artifact was shipped
• What made a past result work
• The workflow around a date, commit, topic, or repository event
• The sessions, tools, skills, or decisions that led to an artifact
• A repeatable recipe based on prior local agent work

Do not use

Do not use this skill for:

• Generic recent activity summaries
• New implementation planning before past evidence is inspected
• Project status reports not tied to a concrete artifact, date, or commit
• Reconstructing work when local ax data is unavailable

Instructions

1. Identify the anchor:
   • Commit SHA: use it directly.
   • Date: use a small window around that date.
   • Topic or artifact name: search commits first.
   • Current repo, recent work: inspect recent sessions for the repo.
2. Resolve candidate sessions:

ax recall "<topic>" --sources=commit --json
ax sessions near <sha> --json
ax sessions around <date> --days=3 --json
ax sessions here --days=14 --json

3. Pick up to five sessions. Prefer sessions with high turn counts, matching commits, related files, or relevant tool and skill usage.
4. Inspect each selected session:

ax sessions show <id> --json
ax sessions show <id> --by-role
ax sessions show <id> --expand=<subagent-uuid>

5. If ax cannot connect to its database, tell the user to start ax services and stop. Do not invent evidence.
6. Write the reconstruction from evidence:
   • Ordered workflow arc: what happened first, next, and last.
   • Skill/tool role: what each important skill or tool contributed.
   • Key decisions: user steering points or tradeoffs that changed the work.
   • Repeatable recipe: how to run a similar workflow again.

Output

Return Markdown with:

• A one-paragraph summary
• An ordered workflow timeline
• Key evidence with session ids, commits, or commands
• A short reproducible recipe when enough evidence exists

Keep the tone factual. Cite local evidence. Mark unknowns instead of filling gaps.