<< All versions

Skill v1.0.2

currentAutomated scan100/100
mxrsv/agents-skills/codebase-onboarding

2 files

──Details
PublishedOctober 22, 2026 at 10:26 PM
Content Hashsha256:1fde167096c78c89...
Git SHA6e5507370440
Bump Typepatch
Compare with v1.0.1
──Files
Files (1 file, 3.2 KB)
SKILL.md3.2 KBactive
SKILL.md · 62 lines · 3.2 KB

version: "1.0.2" name: codebase-onboarding description: Use when entering a new codebase, needing a fast architecture map, or generating onboarding notes from an unfamiliar project. Focuses on reconnaissance first, then targeted reads. Describes the repo for navigation only; to judge architecture, dependency risk or docs drift, use review-health.


Codebase Onboarding

Use this skill for fast orientation in a new repository.

Workflow

  1. Reconnaissance first:
  • inspect top-level structure
  • detect language, framework, package manager, tests, CI, env files
  1. Identify:
  • entry points
  • key directories
  • request/data flow
  • project conventions
  1. Read only the files that matter for those findings.
  2. Emit the result according to the contract in the "Bootstrap" section below.

Output Shape

  • Stack
  • Architecture
  • Key directories
  • Common commands
  • Conventions
  • Gotchas

Rules

  • Do not read everything.
  • Prefer rg, manifests, config files, and a few representative files.
  • Optimize for navigation and execution, not encyclopedic coverage.

Bootstrap (--bootstrap)

Map the recon results to the right files per the D-rules:

Recon itemTarget fileTemplate
Stack, Common commands, Conventions, GotchasAGENTS.md (repo root)~/.claude/templates/AGENTS.template.md
Architecture, request/data flowdocs/internals/overview.md~/.claude/templates/internals-overview.template.md
—docs/README.md — index pointing to user/, internals/, operations/ (omit tiers that do not exist)—

"Current status" belongs in the existing task plan's Handoff section, or in chat for small work (D0/D4). Do not create a separate status document or require a tracker.

Rules when generating:

  • NEVER overwrite an existing file. Already exists → skip it and report "already exists". Fixing an old file to comply with D1/D6 is separate manual work, NOT part of bootstrap.
  • Behavior claims in AGENTS.md MUST be markdown links relative to the file containing the link (D6). In docs/internals/*.md, links out to code → start with ../../; no label needed.
  • docs/internals/overview.md keeps only what "a maintainer would get wrong without it": decisions + rationale, cross-module constraints, traps. No file catalogs, no control-flow retelling (D9).
  • Do not make things up. Cannot verify → write unknown.
  • When done, run bash ~/.claude/scripts/docs-compliance.sh <repo> and bash ~/.claude/scripts/docs-anchors.sh <repo>, and paste the output.

Permissions

Bootstrap authorizes creating the missing files above without overwriting existing ones. Routine documentation needs no separate approval; unclear requirements and material decisions follow D10. Staging and committing follow the task's authorization.

← v1.0.1All versions