agentleFS
adriancoadrianco/retort/skills/run-summary/SKILL.md

run-summary

Skill212 starsChanged 7 months agoApache-2.0View on GitHub

Summarize the architecture of code generated by a single retort run. Produces module-level structure, interfaces, and control flow in a form suitable for cross-run comparison — not a full codebase-summary.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read TASK.md to understand what the code is supposed to do — the "surface" of the project.

    line 25
  2. Very specific:

    If the code has none of the above categories, write (none) under that heading.

    line 77
  3. Very specific:

    Write the index.md with Shape: unparseable — agent output did not produce a valid project.

    line 130
  4. Distinctive:

    Leave the other files near-empty with a one-line explanation.

    line 131

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Run Summary

Overview

Retort runs produce small, single-task codebases (one REST API, one CLI, one service). Full codebase-summary treatment is overkill — but a consistent lightweight summary per run is exactly what makes cross-run comparison useful.

This skill is an intentionally scoped-down adaptation of pourpoise's codebase-summary. It runs in seconds, not minutes, and emits only what's useful when comparing 48+ small generated projects.

Parameters

  • codebase_path (required): The run directory (same as evaluate-run's run_dir)
  • output_dir (optional, default: {codebase_path}/summary): Where to write the summary files

Steps

1. Infer the surface

Read TASK.md to understand what the code is supposed to do — the "surface" of the project. One paragraph, no judgments.

2. Map the modules

Generate {output_dir}/modules.md:

# Modules

| Path | Purpose | Entry points |
|------|---------|--------------|
| src/app.py | HTTP server, route handlers | `app`, `create_app()` |
| src/models.py | SQLAlchemy models | `Book`, `Base` |
| src/db.py | Connection + migrations | `get_engine()` |
| tests/test_app.py | API integration tests | 8 test functions |

Constraints:

  • You MUST list every non-generated source file (skip node_modules, target, __pycache__, .git, lock files, build artifacts).
  • The "Purpose" column is one line extracted from the code, not invented.
  • The "Entry points" column is the publicly-named functions, classes, or exported symbols — not every local helper.

3. Describe the interfaces

Generate {output_dir}/interfaces.md covering what the code exposes:

  • HTTP routes (method, path, short description)
  • CLI commands (subcommand + flags)
  • Library API (exported classes/functions)
  • Data schemas (tables, message formats)

Example:

# Interfaces

## HTTP routes

| Method | Path | Returns | Handler |
|--------|------|---------|---------|
| GET | /books | `[Book]` | `app.py:list_books` |
| POST | /books | `Book` | `app.py:create_book` |
| GET | /books/{id} | `Book \| 404` | `app.py:get_book` |

## Data schema

`books` table: id (int, pk), title (str), author (str), year (int).

Constraints:

  • You MUST grep / static-analyze the code rather than execute it to discover interfaces.
  • You MUST NOT invent endpoints the code doesn't actually declare.
  • If the code has none of the above categories, write (none) under that heading.

4. Trace the dominant control flow

Generate {output_dir}/flow.md with one Mermaid diagram showing the happy-path request/response for the main feature, plus a one-paragraph narration.

# Flow

```mermaid
sequenceDiagram
    Client->>app.py: GET /books
    app.py->>db.py: get_session()
    db.py-->>app.py: Session
    app.py->>Book: query.all()
    Book-->>app.py: [Book]
    app.py-->>Client: 200 {json}

A request to GET /books opens a DB session via db.py:get_session(), queries all Book rows, and returns them as JSON. No pagination, no filtering.


Constraints:
- You MUST pick the single most representative flow — the one a user of the generated code would hit first.
- You MUST note deviations from common patterns ("no input validation", "no error handling", "synchronous DB access in async handler").
- The narration MUST be factual, not prescriptive.

### 5. Write the index

Generate `{output_dir}/index.md` linking to the other three files and giving a 3–4 bullet summary:

```markdown
# Summary: {cell_name} · rep {replicate}

- **Shape:** {one-line description — "Flask REST API with SQLAlchemy", "Go net/http CRUD with in-memory store", etc.}
- **Structure:** {n} modules, {n} test files
- **Interfaces:** {n} HTTP routes / {n} CLI commands / {n} exported functions
- **Notable:** {what stands out — simplest/most-complex approach seen, unusual library choice, etc.}

See [modules.md](modules.md), [interfaces.md](interfaces.md), [flow.md](flow.md).

Constraints Summary

  • You MUST finish in under 90 seconds wall-clock. This is the fast-path summary.
  • You MUST NOT read anything under node_modules/, target/, __pycache__/, .git/, dist/, build/.
  • You MUST write exactly four files: index.md, modules.md, interfaces.md, flow.md. No more, no less.
  • You MUST keep descriptions factual — no quality judgments (that's evaluate-run's job).
  • Output files MUST be valid markdown that renders correctly in GitHub's viewer (Mermaid in a ```mermaid fenced block).

Troubleshooting

Code is unparseable / generated output is garbage

  • Write the index.md with **Shape:** unparseable — agent output did not produce a valid project.
  • Leave the other files near-empty with a one-line explanation.
  • Exit 0 so evaluate-run can still complete.

Too many files to summarize

  • This shouldn't happen in retort's small-task workspaces. If it does, cap the modules table at 50 rows and note the truncation in index.md.

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.