Todo

Maintain the master TODO

Keep the project's docs/TODO.md (the live backlog) and docs/TODO-done.md (the dated archive) accurate, honest, and well-ordered. Template + full legend live in those files' headers.

Status markers (status first, then optional priority)

• [ ] not started · [~] in progress / partial · [?] implemented but unverified (code looks done; not

yet proven). There is no [x] in TODO.md — done+verified items leave the file (see Move rule).

• [P] is a separate priority marker, prepended: [ ] [P] ….
• Verification type on every item: 🤖 = an agent can prove it with a test/command · 🙋 = needs a human

(manual test, judgment, real-world check). Every item carries a → verify: line stating exactly how.

• Exposure tags (optional): 🌍 public-eligible · 🔒 internal-only · ⚖ needs ToS/Privacy review.

What this skill does

• Add an item to the right section with a status marker, verification type + → verify:, deps:, and any

tags. New work without a clear verify is incomplete — ask or infer one.

• Advance status: [ ]→[~]→[?]→ (verified) → moved out. Set [?] when code is written but the verify

hasn't passed yet. Never jump to done on assumption.

• Sort open items by priority → ease → dependency count (quick unblocked wins first, then foundational,

then dependent systems). Keep ## 0. In-flight to what's actively in progress.

• Move on done: when an item is completed and its verify has actually passed (🤖) or its owner signed

off (🙋), move the whole item into docs/TODO-done.md under today's real date (get it from the system — never invent it), preserving the description + the evidence (the command that passed / PR / who verified).

Hard rules

• Do not tick or move an item to done without the verification having actually been observed (constitution

A2). [?] is the honest holding state for "implemented, not proven."

• Default every item to 🤖 (AI-verifiable — the overlord/tester runs the check and reads the output). Make

items auto-verifiable: write the e2e / visual / contract / injected-error test that proves it. Never ask the user to verify a 🤖 item — that's the agent's job (A2).

• Reserve 🙋 for genuine human-acceptance only: a subjective look-and-feel call, or a check needing the

human's credentials / a live deploy (account/login/live-key/deploy work). The autonomous loop does not stop to ask about 🙋 items: it builds everything around them, defers them to docs/DEPLOYMENT.md, and keeps going; they're handed off and run together at the end. The overlord must not mark 🙋 items done on its own.

• Keep it truthful: stalled work is [~] with a note, not a silent [ ].
• Never invent a date; never delete history from TODO-done.md.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: gabrieldabbah
• Source: gabrieldabbah/genesis
• License: MIT
• Homepage: https://brandcrafter.app

Install and usage instructions live in the source repository linked above.