agentleFS
Sign inStart free
JoahanMoralesJoahanMorales/upiixia-agentic-kit/kit/CLAUDE.md

upiixia-agentic-kit / kit

CLAUDE.md12 starsChanged 5 days agoMITView on GitHub

Load .uak/docs/ only when AGENTS.md §7 or your mode says so. Minimal context leaves tokens for the task.

We split this CLAUDE.md into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 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.

  • 4

    Very specific

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

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

    All modes: /uak-setup, /uak-plan, /uak-start, /uak-loop, /uak-advise, /uak-ship, /uak-review, /uak-handoff, /uak-secure, /uak-retro.

    line 16
  2. Very specific:

    Never fan out Opus subagents.

    line 30
  3. Very specific:

    Run tests with bash .uak/bin/q ..., using pytest -q --tb=short -x.

    line 31
  4. Very specific:

    Avoid editing AGENTS.md or CLAUDE.md while agents are running: it invalidates every agent's prompt cache.

    line 32
  5. Distinctive:

    Keep only: ID, branch, worktree, exact next step, pending criteria, failures with exact errors, verify commands, decisions and unanswered …

    line 35

≈ “Common” is an estimate: at least 1 in 5 of the 25,680 CLAUDE.md 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
CLAUDE.md

Shared instructions

@AGENTS.md

Load .uak/docs/ only when AGENTS.md §7 or your mode says so. Minimal context leaves tokens for the task.

Claude Code only

  • Hooks:
    • SessionStart already ran uak next and showed your inbox.
    • PostToolUse sends heartbeats and injects only actionable messages. On Bash, an output over 12k chars reaches you as head + tail (a diff: a per-file index with line ranges) + the path of the full log: read it by ranges (sed -n, grep -n), or rerun with UAK_RAW=1 only if you need everything.
    • Stop won't let you finish with unhandled messages.
    • PreToolUse (guard) blocks destructive or publishing commands, even when chained.
    • UserPromptSubmit blocks pasted secrets.
    • The status line shows task, 5 h/7 d plan usage and uak pace; when the pace level changes, a hook tells you once.
  • Commands:
    • All modes: /uak-setup, /uak-plan, /uak-start, /uak-loop, /uak-advise, /uak-ship, /uak-review, /uak-handoff, /uak-secure, /uak-retro.
    • sprint: /uak-design-lock, /uak-demo.
    • marathon: /uak-spec.
  • Subagents (isolated context, summary only):
    • uak-scout (Haiku): read-only lookups.
    • uak-runner (Haiku): tests, smoke, app.
    • uak-reviewer: reviews.
    • design-critic: sprint UI screenshots.
    • security-reviewer (Opus): marathon sensitive PRs.
    • uak-architect (Opus): checkpoints only.
    • Explore: only when uak-scout is not enough (it runs on the lead model, so it is pricier).
  • Orchestration (.uak/docs/ORCHESTRATION.md):
    • You are the Sonnet lead. Send uak-scout (Haiku) for any lookup needing 3 or more searches, up to 3 in parallel on disjoint questions. Do one-search lookups yourself.
    • Consult the Opus advisor only at the 3 checkpoints: plan lock (more than 2 files or a contract), the same failure twice, done on risky diffs. If the advisor tool is missing or the session is long, use /uak-advise <plan|stuck|done> instead.
    • Never fan out Opus subagents. The guard agent hook caps parallel subagents and the Opus budget per session, scaled by uak pace (surge 2×, conserve ½, critical 1 and no Opus).
  • Run tests with bash .uak/bin/q ..., using pytest -q --tb=short -x. Edit with Edit; never rewrite whole files.
  • Avoid editing AGENTS.md or CLAUDE.md while agents are running: it invalidates every agent's prompt cache. Urgent amendments go through uak decision.

When compacting

Keep only: ID, branch, worktree, exact next step, pending criteria, failures with exact errors, verify commands, decisions and unanswered inbox. Drop command output, diffs and explorations.

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.

Reviews can't be read right now.

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.

Check your setup, free