Skip to content
Agentmart
Submit a tool

review-lessons-reconcile

Cross-check the CURRENT project's mined review lessons against the global rule tree (~/.claude/CLAUDE.md and the reference notes its routing table names in…

Skill by hmdghoul

IndexedApache-2.0
Write a review

Chalk Labs

Need help deploying review-lessons-reconcile?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/hmdghoul-review-lessons-reconcile

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Cross-check the CURRENT project's mined review lessons against the global rule tree (~/.claude/CLAUDE.md and the reference notes its routing table names in the vault) and that project's memory, then report decisions — what is already covered, what should be added or strengthened, where it belongs, and what conflicts with an existing rule. Stops immediately if this repository has no review lessons yet; never reconciles another project's. Proposes only; never edits a rule file or memory without explicit approval. Use after running review-lessons, or when asked whether the lessons are reflected in the rules.

Use cases: documents, productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Nov 1, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/review-lessons-reconcile

What's inside

Skill name
review-lessons-reconcile
Description
Cross-check the CURRENT project's mined review lessons against the global rule tree (~/.claude/CLAUDE.md and the reference notes its routing table names in the vault) and that project's memory, then report decisions — what is already covered, what should be added or strengthened, where it belongs, and what conflicts with an existing rule. Stops immediately if this repository has no review lessons yet; never reconciles another project's. Proposes only; never edits a rule file or memory without explicit approval. Use after running review-lessons, or when asked whether the lessons are reflected in the rules.
License
Apache-2.0
Instructions
70 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Reconcile review lessons with the rules

For each lesson from `review-lessons`, decide whether it is already a rule, should become one, and where it belongs. The output is decisions for the user to approve: **propose, never apply.**

## Step 0: Gate

```bash
gh repo view --json nameWithOwner -q .nameWithOwner        # -> <owner>/<name>
ls ~/.claude/lessons/<owner>/<name>/.review-lessons-state.json
```

If the file doesn't exist, report exactly this and end. Don't use another project's lessons or reconstruct lessons from anything else:

> No review lessons exist for `<owner>/<name>`. Run the `review-lessons` skill for this repository first — this skill only reconciles lessons that were mined from it.

If there is no git repo or `gh` can't resolve a remote, stop the same way. If the state file has zero lessons, stop and say the last run produced none. Otherwise, load its `lessons` array (`mistake`, `whyItMatters`, `whatToCheck`, `occurrences`, `prNumbers`) rather than parsing the markdown.

## Step 1: Read the current rule tree, every time

Never rely on a remembered structure.
- Read `~/.claude/CLAUDE.md`. Its *Reference Notes* table is the authoritative list of rule files. Read every note it lists, in full, from the folder it names. Nothing already in context substitutes for this.
- Read this project's `~/.claude/projects/<project-slug>/memory/MEMORY.md`, plus any memory file whose index line relates to a lesson.
- Read the analysed repo's own `CLAUDE.md` if it has one. It outranks the global defaults, and a lesson it covers needs nothing.

## Step 2: Route each lesson

1. **Universal or repo-specific?** Per *Memory Repos* in `CLAUDE.md`, a lesson that names a table, service, bundle filename or framework only this repo uses is repo-specific, however strong the evidence. It goes to project memory.
2. **Which file and section?** Use the tree's actual headers; this table is only a guide:

| Lesson is about | Target |
|---|---|
| Language-agnostic style, comments, additive/revertible change | `Code Style and Change Conventions` note |
| Matching existing patterns, architecture, layering, configuration, transactions and locks, jobs and events, logging | `Service Architecture Conventions` note |
| Kotlin idioms, Spring, JPA/Hibernate, build | `Kotlin and Spring Conventions` note |
| Queries, schema, migrations | `SQL and Schema Conventions` note |
| How a review finding, an explanation, or ticket text is written | `Review and Ticket Writing` note |
| Process gates (ticket premise, rebasing, commit messages), precedence, shell, workflow, git safety, memory, output | `CLAUDE.md` |
| A fact true only of this repo | that project's memory |

If no existing section fits, say so and propose a new one rather than forcing a half-match.

## Step 3: One verdict per lesson

- **already-covered**: quote the existing line verbatim and name its file. Say this plainly and often.
- **strengthen**: give the current line, the replacement, and what the evidence adds.
- **add**: give the exact line, file and section.
- **add-to-memory**: give the file name, `description`, `metadata.type`, body and `MEMORY.md` pointer line. If an existing memory covers the topic, propose editing it instead.
- **conflict**: the lesson contradicts a rule or a recorded decision, including one in project memory. Show both sides and ask; never resolve it yourself.
- **no-action**: say why (too situational, already enforced by tooling, or a one-off).

## Step 4: Report

Lead with a table of lesson, verdict and target. Then give details in the order a person would act on them: paste-ready text for each proposal, with the occurrence count and 2-3 PR numbers as evidence. Ask which to apply. Apply only approved items, one target at a time, and re-read each file right before editing it.

## Quality bars

- A `CLAUDE.md` line costs every session in every project; a note line costs only the sessions that trigger that note. If a rule fits both, put it in the note. Prefer strengthening an existing line to adding a neighbour, and one precise sentence to three hedged ones.
- Never propose a global rule that only one repo needs; route it to memory.
- Match the target file's voice: terse, imperative, second person, reason inline.
- Quote before claiming "already covered" or "conflicts".
- Don't propose what a formatter, compiler or CI already enforces.
- Evidence strength is not rule-worthiness. 100 occurrences may be a linter's job, and 5 may be a data-corruption class. Say which is which.

## Safety

- Editing `CLAUDE.md`, a vault note or a memory file needs explicit approval in the message that asks for it. Once approved memory content is written, commit it per *Memory Repos*.
- Never edit, stage or commit the `.claude` repo's tracked files as a side effect.
- Apply approved vault edits with the Edit tool only. Never stage, commit or push the vault.

From hmdghoul/claude-code-global-config, Apache-2.0.

More from hmdghoul/claude-code-global-config

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Sep 25, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
Apache-2.0
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from hmdghoul

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get review-lessons-reconcile running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying review-lessons-reconcile?

CallGet help
Cite this page

review-lessons-reconcile skill. Agent Mart. https://agentmart.dev/skills/hmdghoul-review-lessons-reconcile (last updated 2026-10-24)