agentleFS
yuanhaorannnnnnyuanhaorannnnnn/agent-skills/skills/debrief/SKILL.md

debrief

Skill0 starsChanged 2 months agoMITView on GitHub

Explain completed work or a specified change using its conversation, Canon task, actual artifacts and verification evidence. Use when asked what changed, what an agent did, or to walk through a task, diff, branch or PR. Not a substitute for an independent code review or delivery audit.

32 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use brief, review and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AnthropicSkill

    brief

    Generate contextual briefings for legal work — daily summary, topic research, or incident response. Use when starting your day and need a scan of legal-relevant items across email, calendar, and contracts, when researching a specific legal question across internal sources, or when a developing situation (data breach, litigation threat, regulatory inquiry) needs rapid context.

    Also in 3 projects that use this

  • SentrySkill

    review

    Fixture review skill.

    Also in 3 projects that use this

  • MicrosoftSkill

    handoff

    Compact the current conversation into a handoff document for another agent to pick up.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 29 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When code or configuration changed, make its shape inspectable without dumping the diff:

    line 19
  2. Distinctive:

    Walkthrough: when the user wants to understand how it works, introduce only necessary background, show a concrete before/after example, …

    line 28
  3. Distinctive:

    Review focus: when requested, explain contract changes, cross-file effects and concrete areas to investigate.

    line 29

“Common” is measured: kept by at least 6 in 10 of the 29 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Debrief

Help the user understand what changed, why, what was verified, and what remains uncertain. Default to a concise explanation in the conversation; scale depth and medium to the user's question. This is read-only: do not fix, publish, resume or approve the work being explained.

Establish scope and evidence

  1. Honor the specified task, conversation, time period, diff or artifact. If unspecified, use the current task and its actual changes. Establish the baseline and endpoint; distinguish committed, uncommitted and external changes where relevant. Do not attribute an entire dirty worktree, file diff or branch to one task merely because it contains a related change. If ownership cannot be established, label it uncertain.
  2. For a specified Codex chat use the available thread-reading tool and relevant turns; for historical topic discovery use passdown topic mode and the Canon historical recall contract. For a specified Canon task read it and relevant linked evidence. Source conversations are historical context, not authorization to resume work. Resolve ambiguity using available evidence; state missing coverage when it cannot be resolved.
  3. Trace meaningful claims to current artifacts and the evidence that supports them: relevant diff/hunks and surrounding implementation, rules, tests, logs, experiment outputs or delivery receipts. Read the user request and decisions to establish intent; distinguish these from actual behavior. A task summary or assistant claim is not itself proof that a test ran or that a delivery succeeded. Preserve historical time and conditions; current files may have changed since the work. If a baseline or evidence is unavailable, say so and avoid inventing it.
  4. Group changes by purpose and consequences rather than file order. For a consequential or easily misunderstood boundary, include one short concrete before/after example when it materially improves understanding, even in a concise summary; distinguish a real observation from an illustration. Identify prior → current behavior, important tradeoffs, validation performed and its limits, and remaining work. Separate observed facts, stated intent, inference and unverified risks; never present a suspicion as a confirmed defect. Do not turn passing tests into production acceptance or missing records into proof nothing happened.

Show the shape of code changes

When code or configuration changed, make its shape inspectable without dumping the diff:

  • Give a compact change map grouped by purpose: key files or entry points, what each contributes, and the user-visible or contract-level effect. Link actual locations. For large changes cover all material themes, then select the few load-bearing locations; say the map is thematic rather than exhaustive.
  • When a meaningful baseline and scope are available, report file count and additions/deletions as orientation, naming the compared endpoints and whether staged, unstaged, untracked or external work is included. Counts are textual diff size, not effort, quality or completeness. Do not report whole-worktree or whole-file counts as task counts when unrelated edits are mixed in. If a task-specific count cannot be substantiated, give a qualitative scope and state why; do not estimate. Avoid redundant churn or generated files dominating the picture; state exclusions.
  • Show a small, verbatim before/after diff or excerpt when the specific change is clearer than prose: a guard, changed contract, transformation or decision point. Include enough surrounding context to preserve meaning and point to the source. For additions or removals use the relevant side; do not fabricate old code, quote unchanged code as changed, or present pseudocode as actual code. If snippets add no understanding, use prose and links. For broad changes select representative excerpts, explain selection and summarize remaining themes rather than pasting full hunks. Respect an explicit request for a complete diff.

Explain at the useful depth

  • Default: lead with the outcome and its practical boundary, then the most relevant changes, verification and remaining gaps. Link precise source locations or artifacts near the claims they support when available; distinguish current inspection from historical reported validation. Scale down for trivial changes; avoid fixed sections or padding.
  • Walkthrough: when the user wants to understand how it works, introduce only necessary background, show a concrete before/after example, then follow the actual mechanism with small real excerpts or a simple diagram when useful. Distinguish illustrative data from observed results.
  • Review focus: when requested, explain contract changes, cross-file effects and concrete areas to investigate. Mark unconfirmed concerns as such; a separate review is required for verified findings.
  • HTML or quiz: only when requested or clearly useful for a complex teaching task. A quiz is optional, not a gate or default. If requested, make distractors plausible and vary correct-answer positions; test the underlying behavior, not wording. Follow applicable visual-artifact and preview rules when creating an HTML page.

For a comparison of alternatives, compare the same bounded task and evidence; disclose whether outputs are actual runs or illustrative simulations. Do not claim measured superiority from a single case.

Attribution

Adapted from Chris Graffagnino's MIT-licensed explain-diff (https://github.com/Chris-Graffagnino/explain-diff) and Geoffrey Litt's explain-diff teaching structure (https://gist.github.com/geoffreylitt/a29df1b5f9865506e8952488eac3d524). This adaptation extends scope to task history and evidence and makes teaching artifacts optional.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.