Extract learnings from AI agent sessions into reusable fixes

This skill reconstructs the workflow behind a past coding-agent artifact - a shipped feature, PR, demo, refactor, or report - by using the local ax graph to connect commits, sessions, turns, skills, and tool traces into a short "how this got made" narrative. It works in four steps: resolve an anchor (commit SHA, date, feature name, or "this repo recently"), pick the few relevant sessions, inspect the session trail for skills used, steering points, files/tests/commands that changed direction, and verification steps, then write a short evidence-grounded reconstruction with an anchor, a 4-8 step ordered workflow, key decisions, verification, and a compact reproducer brief, citing session IDs, commit SHAs, and file paths where available.

Use it when the user asks "how did we build X?", "what made X work?", or "extract the workflow behind this artifact", when the anchor is a commit SHA, date, feature name, PR, session, or repo-local artifact, or when the user wants the sequence of agent skills, prompts, commands, decisions, and checks that led to a result. Do not use it for a generic activity summary - use normal session listing instead. It reconstructs workflow, not correctness, so the code and project checks still need inspecting separately before making engineering decisions.

Input is an anchor - a commit SHA, date or date range, topic/feature/artifact name, or "this repo recently" - resolved via read-only ax commands, for example:

ax recall "live ingest dashboard" --sources=turn,commit,skill --scope=here

Output is an inline (or, on request, file) reconstruction: the resolved anchor, an ordered 4-8 step workflow, key decisions, verification evidence, and a reproducer brief, citing session IDs, commit SHAs, and file paths.

It requires a working local ax installation on PATH with a reachable local ax database; if ax cannot connect, the skill reports the connection failure and stops rather than guessing from memory. It only sees sessions, commits, skills, and tool traces that ax has already ingested, and session data can be incomplete when an agent provider omits tool output, cost, or reasoning data. Related skills: agenttrace-session-audit for session health/cost/latency/tool-failure audits, domain-modeling when the reconstruction reveals terminology or architecture worth capturing, and planning-with-files for turning the reconstructed recipe into a tracked plan.

Engineers and agents who need to explain or reproduce how a past artifact was actually built, working strictly from ax's own read-only inspection commands - never inventing missing skills, costs, commands, or decisions, and never uploading private transcripts, session logs, prompts, tool outputs, or local database exports. Secrets, tokens, customer data, and private conversation text are redacted from any summary, and the skill never mutates .ax/, regenerates indexes, publishes reports, or alters repositories as part of reconstruction.