skill-debrief
- Repo stars 435
- Author updated Live
- Author repo agr
- Domain
- Other
- Compatible agents
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- Trust score
- 88 / 100 · community maintained
- Author / version / license
- @computerlovetech · no license declared
- Token usage
- Lean
- Setup complexity
- Plug-and-play
- External API key
- Not required
- Operating systems
- Unspecified (assume cross-platform)
- Runtime requirements
- No special requirements
- Permissions
-
- Read-only
- Write / modify
- Network behavior
- Local-only
- Install commands
- 26 variants
Profile is derived at build time from SKILL.md and install vectors. Subject to drift from author intent.
Heads up: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: skill-debrief
description: > Capture lessons from a session into the skill that drove it. The default shape is listen → pro…
category: other
runtime: no special runtime
---
# skill-debrief output preview
## PART A: Task fit
- Use case: > Capture lessons from a session into the skill that drove it. The default shape is listen → propose → align → apply → re-install. Trigger when the user wants to debrief an existing SKILL.md based on what runs entirely locally. Works with Claude Code, Cursor, Cline and 23 more..
- Inputs: target material, constraints, expected output, and acceptance criteria.
- Evidence boundary: follow “When to use / Step 1: Identify the skill / Step 2: Receive feedback” and do not present inference as author intent.
## PART B: Execution result
- **01** The card summarizes the use case; runtime output centers on “> Capture lessons from a session into the skill that drove it. The default shape is listen → propose → align → apply → re-install. Trigger when the user wants to debrief an existing SKILL.md based on what runs entirely locally. Works with Claude Code, Cursor, Cline and 23 more.”.
- **02** When the source has headings, the agent prioritizes “When to use / Step 1: Identify the skill / Step 2: Receive feedback” so the result follows the author’s structure.
- **03** Typical output includes task judgment, concrete steps, required commands or file edits, validation, and follow-up options.
- **04** Risk context follows the fingerprint: read files, write/modify files; mostly runs locally; usually needs no extra API key.
## Running Rules
- read files, write/modify files; mostly runs locally; usually needs no extra API key.
- Validate with a small sample before expanding scope.
- Return the result, validation criteria, and next iteration options. The source mentions slash commands such as `/tmp`; use them first when your agent supports command triggers.
Name target files or source material, expected output, forbidden changes, and whether network or shell access is allowed. Permission fingerprint: read files, write/modify files.
Start with a small task and check whether the result follows “When to use / Step 1: Identify the skill / Step 2: Receive feedback”. Inspect diffs, logs, previews, or tests before expanding scope.
Confirm the final output includes a concrete result, evidence, and next action. If it stays generic, tighten inputs, boundaries, and acceptance criteria.
---
name: skill-debrief
description: > Capture lessons from a session into the skill that drove it. The default shape is listen → pro…
category: other
source: computerlovetech/agr
---
# skill-debrief
## When to use
- > Capture lessons from a session into the skill that drove it. The default shape is listen → propose → align → apply →…
- Use it when the task has clear inputs, repeatable steps, and validation criteria.
## What to provide
- Target material, scope, expected result, and forbidden changes.
- Whether network, commands, file writes, or external services are allowed.
## Execution rules
- Organize steps around “When to use / Step 1: Identify the skill / Step 2: Receive feedback” and keep inference separate from source facts.
- read files, write/modify files; mostly runs locally; usually needs no extra API key.
- Validate with a small sample before expanding the task.
## Output requirements
- Return the deliverable, key evidence, validation method, and next action.
- Mark missing information as unknown; do not invent commands, platforms, or dependencies. The author source anchors workflow facts; repository files anchor sources and commands; Fluxly only adds fit, limitations, and quality judgment.
skill "skill-debrief" {
input -> user goal + target files + boundaries + acceptance criteria
context -> When to use / Step 1: Identify the skill / Step 2: Receive feedback
rules -> SKILL.md triggers / order / output contract
runtime -> no special runtime | read files, write/modify files | mostly runs locally
guardrails -> usually needs no extra API key + small-sample validation + diff/log review
output -> copyable result + checklist + next iteration
} Skill Debrief
Capture lessons from a session into the skill that drove it. The default shape is listen → propose → align → apply → re-install.
When to use
Trigger when the user wants to debrief an existing SKILL.md based on what happened in the session. Examples:
- "debrief the X skill" / "let's debrief X"
- "retrospective on X" / "feedback on X"
- "improve the X skill" / "let's update X based on what we learned"
- "X skill should also handle …"
- "X didn't trigger when it should have"
Do NOT use this skill for:
- Greenfield skill authoring. Use
agr initto scaffold a SKILL.md and defer the body content to the user — or to a dedicated authoring skill such asanthropics/skills/skill-creator(agr add anthropics/skills/skill-creator). - Installing / syncing / removing skills. That's plain
agrCLI work (agr add,agr sync,agr upgrade,agr remove).
Step 1: Identify the skill
Ask which skill is being improved if it isn't obvious from context. Then locate the source:
agr list # see installed deps and short names
ls skills/ # in-repo source if present
cat agr.toml # see whether the dep is local-path or remote
Two cases — they have different update paths:
| Case | Source location | Update path |
|---|---|---|
In-repo ({path = "./skills/<name>", type = "skill"} in agr.toml) |
skills/<name>/ |
Edit source → commit → agr upgrade <name> |
Remote ({handle = "user/repo/<name>", …}) |
Upstream GitHub repo | Cannot edit directly — see Step 5 |
If the skill isn't installed at all but the user wants to improve it, ask whether to add it first (and which case applies).
Step 2: Receive feedback
Listen. The user invoked this skill because they have something to say — let them say it. Do not interrogate. Do not run a checklist of questions at them. Take in whatever they offer, in whatever shape they offer it.
Only ask a clarifying question if you genuinely cannot proceed without one (e.g. the user named a skill that doesn't exist, or two skills share the name and you need to disambiguate). Even then, ask the minimum.
Be dynamic. The user may surface things in any shape — a single
sentence ("the description should also fire on X"), a structured list, or a
ramble that you need to distill. They may also surface things outside the
standard buckets below (rename a section, restructure references/, change
output format, drop a deprecated workflow, fix a typo). Apply whatever the
user actually says.
The buckets below are a mental map for you when distilling what you heard, not a checklist to recite at the user:
description/ triggers — under-fired or over-fired- Gotchas / boundaries — a foot-gun the skill didn't warn about
- Workflow steps — missing, wrong, or out of order
- References — a topic kept needing more depth → new
references/<topic>.md - Examples / output format — vague where it should be concrete
- Pruning — outdated content that misled
Step 3: Propose changes
Summarize what you heard, then propose specific edits. Format:
Proposed changes to
skills/<name>/SKILL.md(and any references):
- Description — add trigger phrase "…" (because: …)
- Boundaries — add: never X (because: discovered this in session)
- New section "Y" — describes the workflow that was missing
Want me to apply these, revise, or add more?
For small edits, show the exact diff inline. For larger changes, summarize first and apply section by section.
Wait for explicit user approval before editing. "yes" / "go ahead" / similar. If the user revises, loop back to Step 2 or 3.
Step 4: Apply (in-repo case)
Edit the source file(s) under skills/<name>/. Then:
git status # confirm only the intended files changed
git add skills/<name>/
git commit -m "skill(<name>): <one-line summary>"
agr upgrade <name> # re-installs into all configured tools, refreshes agr.lock
git add agr.lock
git commit --amend --no-edit # or commit separately; match the repo's style
Do not push. Stop after the commit and let the user push when ready.
Commit message style
Use a conventional-commits-style scope:
skill(<name>): <imperative summary>
Examples:
skill(agr-cli): clarify upgrade vs sync for local paths
skill(agr-cli): add gotcha for same-repo siblings
skill(code-review): drop outdated linter pre-check
If the repo's commit style differs (check git log --oneline -20), match it.
Why agr upgrade and not agr sync?
agr sync only installs missing deps — it does NOT re-copy a local-path
skill that's already installed. agr upgrade <name> re-copies it and
refreshes agr.lock. Use agr upgrade.
(Equivalent: agr add ./skills/<name> --overwrite. Pick upgrade for
consistency — it's the same verb used to refresh remote skills.)
Step 5: Apply (remote case)
If the skill is a remote dep, the change cannot be applied directly. Ask the user which path they want:
Option A — File an issue upstream via gh
Best when the change benefits everyone (a real bug or universal improvement in someone else's published skill).
Resolve the upstream repo from the handle:
anthropics/skills/pdf→--repo anthropics/skillsuser/myrepo/skill→--repo user/myrepo
Confirm the title and body with the user, then:
gh issue create \
--repo <owner>/<repo> \
--title "[<skill-name>] <short summary>" \
--body "$(cat <<'EOF'
## What I observed
<concrete scenario from the session>
## Suggested change
<proposed wording or workflow>
## Why
<reasoning>
EOF
)"
gh issue create posts publicly — treat it the same as any other shared-
state action. Always confirm before running.
If the user has push access and a local clone, also offer to open a PR instead of (or alongside) the issue.
Option B — Fork to in-repo
Best when the change is project-specific or unlikely to be accepted
upstream. Copies the skill into skills/<name>/ so future retros work the
in-repo way.
Suggested flow:
# 1. Find the upstream commit (agr.lock has it)
agr list
# 2. Sparse-checkout or full clone, then copy the folder:
mkdir -p skills
cp -r /tmp/upstream-clone/<skill-folder> skills/<name>
# 3. Swap the dep
agr remove anthropics/skills/<name>
agr add ./skills/<name>
Tell the user this forks the skill — they're now responsible for keeping it current with upstream. Then continue from Step 3 with the in-repo flow.
Both A and B
Offer Option A first when the change is generally useful. Offer Option B when upstream is unlikely to accept, or when the user wants the change now without waiting on upstream.
Step 6: Verify
After re-installing (in-repo case):
agr list # status should be `installed`
diff skills/<name>/SKILL.md .claude/skills/<name>/SKILL.md # should be empty
For remote case (issue filed): confirm the issue URL with the user.
Tell the user what's done and what's pending (commit done, push pending; or issue filed, awaiting response).
Boundaries
- Don't edit a skill without explicit user approval of the proposed changes. Skills are user-owned content — never silently revise.
- Don't push. Commit only; the user pushes when ready.
- Don't open issues or PRs without confirming the title and body with
the user first.
gh issue createis publicly visible — treat it as shared-state. - Don't edit
agr.lockby hand —agr upgraderegenerates it. - Don't broaden the scope of the edit beyond what was discussed. If the user asked to fix one gotcha, don't also restructure the file.
- Don't write a skill from scratch — that's a separate workflow.
See also
anthropics/skills/skill-creator— canonical greenfield skill authoring (install withagr add anthropics/skills/skill-creatorif needed)
Decide Fit First
Design Intent
How To Use It
Boundaries And Review