| name | summary-my-repo |
| description | Use when the user wants an internal repository summary that captures architecture, directory responsibilities, core code paths, and code-backed logic evidence so other engineers can understand the project quickly. |
summary-my-repo
Purpose
Turn a repository into an internal engineering summary pack.
This skill is for fast project onboarding, architecture digestion, and codebase orientation. It should explain the structure and logic of the repo without collapsing into a shallow tree dump.
Use This Skill When
- the user wants to summarize a repository for future collaborators
- the user wants architecture notes plus a readable directory map
- the user wants the most important files and logic paths explained
- the user wants a markdown pack written into the repository for later reuse
Do Not Use This Skill When
- the user wants outward-facing product packaging, investor messaging, or GTM framing
- the user only wants a code review or bug fix
- the repository is so incomplete that core logic cannot be identified from evidence
Non-Negotiable Rules
- Do not stop at listing directories. Explain responsibility boundaries.
- Do not stop at naming files. Explain what each critical file actually does.
- Do not claim architecture that the code does not support.
- Clearly separate
implemented, inferred, planned, and risk.
- Prefer a small number of dense, high-signal markdown files over one giant unfocused document.
- Preserve important logic even when compressing. Omit repetition, not substance.
- Include necessary core code evidence. Pure prose-only packs are incomplete.
- Every core workflow claim must be backed by at least one code snippet or config snippet from the repository.
- For every snippet, include file path and line range.
- Keep snippets focused. Prefer 8-40 lines per snippet over full-file dumps.
Workflow
1. Establish the repo shape
Determine:
- project name
- repo purpose
- top-level directory roles
- main entry docs
- core scripts, services, packages, or applications
- validation or CI shape
2. Prepare the output folder
Before writing the final answer, create or reuse the repo-root output directory:
- root output directory:
summary-my-repo/
- run output directory:
summary-my-repo/YYYY-MM-DD-vN/
Default output files:
summary-my-repo/YYYY-MM-DD-vN/00-overview.md
summary-my-repo/YYYY-MM-DD-vN/01-directory-map.md
summary-my-repo/YYYY-MM-DD-vN/02-core-logic.md
summary-my-repo/YYYY-MM-DD-vN/03-code-evidence.md
Versioning rule:
YYYY-MM-DD uses the current local date of the run
vN starts at v1
- if
summary-my-repo/YYYY-MM-DD-v1/ already exists, increment to v2, then v3, and so on
3. Read docs first, then verify with code
Start with user-facing docs, maintainer docs, and design notes. Then verify claims with:
- directory structure
- core implementation files
- scripts and automation
- registry or manifest files
- tests and CI
- example artifacts when they meaningfully reveal architecture
4. Build the summary pack
Write at least these sections across one or more markdown files:
- repo purpose and current maturity
- architecture snapshot
- directory map with responsibilities
- core workflows and control flow
- key files and why they matter
- invariants, assumptions, and coupling points
- risks, gaps, or confusing areas
- recommended reading order
5. Extract code evidence for core logic
For each core workflow or critical file, capture one or more snippets that prove the behavior:
- shell scripts, language source, config, workflow YAML, or manifest JSON are all valid evidence
- keep each snippet small enough to read quickly
- include only code that directly supports the claim
- avoid copying entire files unless the repo is tiny
6. Emphasize core logic with evidence
For the most important files, explain:
- role in the system
- main inputs
- main outputs
- downstream effects
- why the file is central to understanding the repo
- which snippet ids in
03-code-evidence.md prove each claim
When code contains meaningful control flow, summarize the path step by step and attach the matching snippet ids.
Code Evidence Format (Required)
For each code evidence entry, include:
Snippet ID: stable local id such as S01
File: repository-relative path with line range, for example scripts/validate-repo.sh:24-47
Claim: one sentence describing what this snippet proves
Code: fenced code block copied from the repository
Interpretation: short explanation of how the code supports the claim
Evidence density rule:
- non-trivial repositories: at least 3 snippet entries
- medium or large repositories: usually 5-10 snippet entries
- if fewer than 3 are included, explain why in
00-overview.md
Output Standard
Unless the user asks for a different shape, write a markdown pack to summary-my-repo/YYYY-MM-DD-vN/ using:
templates/00-executive-summary.md
templates/01-directory-map.md
templates/02-core-logic.md
templates/03-code-evidence.md
Output intent by file:
00-overview.md: context, architecture snapshot, workflows, risks, and reading order
01-directory-map.md: annotated tree and ownership boundaries
02-core-logic.md: logic walkthroughs that reference snippet ids from 03-code-evidence.md
03-code-evidence.md: code-first proof layer for the main claims
Read rubrics/coverage-checklist.md before finalizing to ensure the pack covers architecture, directories, workflows, invariants, risks, and code evidence.
If the user also wants an inline answer, provide only a short summary and point to the generated files.