Skip to content

TsukiNi22/ai-utils / skills/ai/onboard

onboard

Verified

Onboard: the context of a project, before working on it

Read everything that says what the project is, where it stands and what earlier sessions already did, then hold it as the context of this conversation. Read-only: nothing built, changed, committed, stopped or resumed.

0. Scope

  • The folder given, else the cwd. Inside a git repository: its root (a subfolder of a monorepo: that part, plus the root files that govern it - README, CLAUDE.md, build, CI).
  • Huge (thousands of files): the structure and the central parts read fully, the rest skimmed; propose a fan-out to subagents before using one (say how many and why), never on your own.

1. The project

  • Tree: git ls-files (else find, skipping .git, build*, node_modules, .venv, dist, caches), sizes, languages.
  • Docs, read entirely: README*, CLAUDE.md / AGENTS.md / .cursorrules / .github/copilot-instructions.md (rules to follow from now on), CONTRIBUTING, CHANGELOG, docs/, wiki pages in the repo, TODO / NOTES / PASSATION.md / handover files.
  • License and credits: LICENSE / COPYING / NOTICE, SPDX headers and file banners, third-party credits, the licenses of the dependencies when they matter.
  • Build and config: CMakeLists / Makefile / package.json / pyproject / Cargo.toml / go.mod..., setup / install scripts, CI workflows (.github/workflows, .gitlab-ci.yml), Dockerfiles, config templates, .gitignore, pre-commit hooks.
  • Code: entry points and the module layout first, then every file (a big one: its structure and the parts that matter); the tests (what they cover, how they run); TODO / FIXME / XXX / HACK (grep, with their place).
  • Links: the remotes, homepage, issues, CI badges, docs sites, submodules, dependencies. With a GitHub remote and gh: gh repo view, gh issue list, gh pr list, gh release list (read-only); other URLs only when they matter.
  • Secrets: never copied - .env, keys, tokens, credentials: their existence and place only.

2. Its history (git)

  • git log (subjects of everything, --stat of the recent ones), the conventions of the messages, the authors, the pace.
  • Branches (local and remote) and how far each is from the main one; tags / releases; worktrees (git worktree list); stashes (git stash list, never popped); submodules.
  • The work not in the history: git status, git diff --stat (and the diff of what matters), unpushed commits, untracked files that look like work in progress.

3. Earlier sessions on this computer

Search by the absolute path of the project (and its parents / subfolders / worktrees), newest first:

  • Claude Code: ~/.claude/projects/<the path, every / and . turned into ->/*.jsonl (worktrees: <repo>--claude-worktrees-<name>), and its memory/ (MEMORY.md and the files it lists); ~/.claude/todos, plans.
  • claude-cluster: claude-cluster list --rtk and claude-cluster external --rtk (every instance: instances/* in ~/.local/share/claude-cluster), the sessions whose folder is the project, their transcripts (logs/), tasks, schedules and todo lists; a running session is only read, never sent anything.
  • Other agents: Codex ~/.codex/sessions, Qwen Code ~/.qwen/projects, opencode opencode session list, Gemini ~/.gemini - grep the project path.
  • Handover files anywhere in the project or its parent: PASSATION.md, HANDOFF.md, NOTES.md.
  • From each conversation: the requests of the user (their words for the important ones), what was done, what was left or failed, the decisions and refusals, the rules given. A long transcript is never dumped: extract the user messages, the last assistant messages and the errors (jq / a short python script), then read what matters.

4. Merge and check

  • Every claim of a session or a doc checked against the code and git ("done" in a transcript but absent from the code: not done). Contradictions: the code and the history win - say so.
  • Sort into: what the project is; done; in progress (with the files touched); left to do; blocked; the decisions; the rules of the user (git, style, language, what never to do); the traps met (flaky tests, quotas, tools missing).

5. Report

  • To the user, terse (rtk): the project in two lines, its state (branch, last commit, pending work), what earlier sessions did and left (dated), the rules to follow, then the possible next steps - the most useful first. Open questions only when something really cannot be settled from what was read.
  • Kept in this conversation as its context. A file (CONTEXT.md, PASSATION.md) only when the user asks; facts worth keeping across sessions offered for the memory, never written without being asked.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 197kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall