A Claude Skill for running a structured after-action review on an AI agent's work, and producing a record you can keep.
The records help you analyze a run, track decisions across agent instances, and stay oriented when you're working with several agents at once.
When you ask Claude to review what an agent did, this skill runs a six-part interview — Intent, Action, Judgment, Deviation, Consequence, Change — and writes the answers into a single comparable artifact: an Agent Action Record. It's manager-led: it interviews the person responsible for the agent and pulls in agent logs or outputs where they're available.
The review surfaces the things that normally vanish the moment an agent finishes: the decisions it made on its own, the moments a human should have been in the loop, the places it was confidently wrong, and what that cost or risked.
SKILL.md— the methodology Claude follows to run the reviewrecord-template.md— the record it writes into (also usable on its own, by hand, with no AI)
Drop this folder into your Claude environment's skills directory. Claude will trigger it automatically when you ask to review, debrief, or investigate what an agent did — e.g. "run an after-action on this agent run" or "what did the agent actually decide here?"
For where skills live and how they load, see the Agent Skills docs.
record-template.md is a plain fill-in form. You can run the same review by hand — work through the six sections in order, write down the answers, keep the record. The records are most useful read across many runs: patterns in the Deviation and Change sections over time tell you more than any single review.
🪂 Shared under the Parachute Commons License 2.0. Read, share, adapt, and
build on it for free with attribution; added terms apply to commercial or
institutional use at scale. See LICENSE.
Source: https://github.com/quarterback/parachute-commons-license