AGENT SKILL

summarize

Produces a structured execution summary of the current session — every action taken, issues found with their root causes, the solutions implemented, verification evidence, and outstanding work. Use when the user types /summarize, asks for a session recap, wants a handoff document, asks "what did we do", or needs a record of changes before ending a session.

License 许可
MIT
最近更新
2026-10-23
Stars
★ 5
Forks
0

01 前言 Front Matter

name
summarize
description
Produces a structured execution summary of the current session — every action taken, issues found with their root causes, the solutions implemented, verification evidence, and outstanding work. Use when the user types /summarize, asks for a session recap, wants a handoff document, asks "what did we do", or needs a record of changes before ending a session.
allowed-tools
Read Grep Glob Bash

03 SKILL.md 正文

Summarize

Produce an accurate, scannable record of what happened in this session.

The summary serves two readers: the user, who wants to confirm what changed, and a future session or teammate, who wants to continue the work without re-deriving it. Write for both.

The Integrity Rule

Report only what actually happened in this session.

  • Do not claim a file was changed if it was not.
  • Do not claim a test passed unless you saw it pass.
  • Do not present a plan as a completed action.
  • Do not omit a failure, a revert, or a dead end. A summary that hides the failed attempt makes the next session repeat it.
  • If you do not know whether something succeeded, say so.

Verify claims against the session record rather than recalling them. Use the tools to confirm file state and repository state.

git status --short
git diff --stat
git log --oneline -10

If the session made no repository changes, say that plainly rather than padding the report.

Structure

Produce these sections in this order. Omit a section only when it is genuinely empty — and when you omit one, say so, because a missing section reads as an oversight.

1. Overview

Two to four sentences. What was the session about, what was the outcome, and what is the state now. A reader who stops here should know whether the work is done.

2. Actions Taken

Every action, in the order it happened, with the evidence.

# Action Target Result
1 Read src/auth/session.ts Located the token refresh path
2 Edit src/auth/session.ts:88 Added expiry check before refresh
3 Run bun test src/auth 12 passed, 0 failed
4 Run bunx tsc --noEmit 2 errors, both pre-existing

Group by type when the list is long: Files created, Files modified, Files deleted, Commands executed, Dependencies changed, Configuration changed. For each modified file, state what changed and why — not just that it changed.

3. Issues and Root Causes

For each problem encountered — reported by the user or discovered during the work.

#### Issue 1 — <one-line symptom>

**Symptom** What was observed, with the exact error text if there was one.

**Root cause** Why it happened. Not the surface location — the actual mechanism.

**Evidence** The file, line, command output, or stack frame that proves the cause.

**Fix** What was changed and why that addresses the cause rather than the symptom.

**Verification** How it was confirmed fixed, with the result.

Distinguish clearly between:

  • Reported by the user — a bug or request they raised.
  • Discovered during work — something you found while doing something else.
  • Pre-existing — a defect you did not introduce and did not fix. Always call these out separately. Conflating a pre-existing problem with your own change makes the diff look worse than it is and hides real work the user still owes.

4. Changes Made

For each change, the substance — not the diff.

  • What the code did before and what it does now.
  • The key logic or structural decision, and the constraint that drove it.
  • Edge cases handled, and edge cases deliberately not handled.
  • Anything a reviewer must know to read the diff correctly.

For an architectural change, describe the before and after shape in a sentence or a small diagram.

5. Verification

What was actually run, and what it returned. This section is where trust is earned or lost.

Check Command Result
Type check bunx tsc --noEmit Pass — 0 errors
Unit tests bun test Pass — 47 passed, 0 failed
Lint bun lint Pass
Build bun build Pass
Manual Loaded /login, submitted bad creds Correct error shown

If a check was not run, list it under Not Verified with the reason. An unverified claim stated as verified is the single most damaging thing a summary can do.

6. Not Verified / Known Gaps

Be specific and unflinching.

  • What was not tested and why (no test exists, no fixture, requires a live service).
  • Assumptions made that were not confirmed.
  • Behavior you changed that has no test coverage.
  • Anything you are uncertain about.

7. Follow-Up

Concrete, ordered next steps. Each names what to do and why it matters.

  1. Add a regression test for the expiry boundary at src/auth/session.ts:88 — the fix is untested, so it can silently regress.
  2. Migrate the two callers of the deprecated helper — they still use the old signature.
  3. Pre-existing: tsc reports 2 errors in src/legacy/* — unrelated to this session, still open.

Separate blocking from optional. Do not pad this list with speculative work.

Output Rules

  • Lead with the outcome. The first line answers "did it work".
  • Use tables for anything enumerable. Actions, files, verification results.
  • Cite locations. src/auth/session.ts:88, not "the session file".
  • Quote exact errors and exact command output. Paraphrasing an error loses the string the reader needs to search for.
  • Distinguish done from planned from blocked. Use explicit labels.
  • Be concise but complete. Cut adjectives, not facts. No "successfully", no "carefully", no "it's worth noting".
  • Keep it readable. A reader should be able to skim the headings and tables and get the shape in ten seconds, then read one section for detail.
  • No emoji, no decorative formatting. This is a record.

Handling an Empty or Trivial Session

If the session was a question, a review with no edits, or an exploration, do not manufacture sections. Report what happened in the honest shape:

## Overview

Reviewed the proposed caching layer. No code changed — the design has a
correctness problem that should be resolved first.

## Findings

| #   | Finding                                                 | Severity | Location             |
| --- | ------------------------------------------------------- | -------- | -------------------- |
| 1   | Cache key omits the tenant id, so tenants share entries | Critical | src/cache/keys.ts:14 |

## Actions Taken

Read-only. No files modified, no commands run beyond search.

## Recommendation

Fix the cache key before implementing. See finding 1.

A short accurate summary beats a long padded one.

Handoff Form

When the user wants to hand off to another session or teammate, add a section that a fresh agent can act on without reading the transcript:

## Handoff

**State** Feature implemented and tested; not yet committed.
**Branch** `feat/session-expiry` — 3 commits ahead of `main`
**Entry point** `src/auth/session.ts` — `refreshSession()` is the changed function
**How to verify** `bun test src/auth && bun run dev`, then load `/login`
**Open** Two callers still use the old signature — see Follow-Up 2.
**Do not** Revert the `!` on the schema change; it is intentional.

Common Mistakes

Mistake Why It Breaks Correct Approach
Claiming tests passed without running them Destroys trust the moment it is checked Run them, or list under Not Verified
Mixing pre-existing defects with your changes Makes the diff look worse and hides the user's remaining work Separate "pre-existing" explicitly
Listing files without saying what changed Unreadable; the user must open every file One line per file describing the change
Paraphrasing an error message The reader cannot search for it Quote it exactly
Reporting a plan as done The next session builds on something that does not exist Label done / planned / blocked
Omitting the failed attempt The next session repeats it Record dead ends and why they failed
Padding with "successfully" and "carefully" Buries the facts Cut the adverbs
No verification section The reader cannot tell whether anything was checked Always state what was run and what it returned
Writing prose where a table belongs Slow to scan Tables for enumerables
Burying the outcome at the end The reader must read everything to learn whether it worked Lead with the outcome
Inventing follow-up work Wastes the reader's attention Only real, actionable next steps
Manufacturing sections for a trivial session Signals padding and undermines the whole report Report the honest shape, however short

04 同仓库更多 Skills

Marvellbrazil
Marvellbrazil/my-agentic-skills

writing-guidelines

Review docs/prose for Writing Guidelines compliance. Use when asked to "review my docs", "check writing style", "audit prose", "review docs voice and tone", or "check this page against the writing handbook".

★ 5
Marvellbrazil
Marvellbrazil/my-agentic-skills

writing-for-agents

Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.

★ 5
Marvellbrazil
Marvellbrazil/my-agentic-skills

wrangler

Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.

★ 5
Marvellbrazil
Marvellbrazil/my-agentic-skills

workers-best-practices

Reviews and authors Cloudflare Workers code against production best practices. Load when writing new Workers, reviewing Worker code, configuring wrangler.jsonc, or checking for common Workers anti-patterns (streaming, floating promises, global state, secrets, bindings, observability). Biases towards retrieval from Cloudflare docs over pre-trained knowledge.

★ 5
Marvellbrazil
Marvellbrazil/my-agentic-skills

web-perf

Analyzes web performance using Chrome DevTools MCP. Measures Core Web Vitals (LCP, INP, CLS) and supplementary metrics (FCP, TBT, Speed Index), identifies render-blocking resources, network dependency chains, layout shifts, caching issues, and accessibility gaps. Use when asked to audit, profile, debug, or optimize page load performance, Lighthouse scores, or site speed. Biases towards retrieval from current documentation over pre-trained knowledge.

★ 5
Marvellbrazil
Marvellbrazil/my-agentic-skills

using-agent-skills

Discovers and invokes agent skills. Use when starting a session, or when you need to decide which skill or workflow applies to the piece of work at hand. This is the meta-skill that governs how all other skills are discovered and invoked.

★ 5