agentleFS
Fadi88Fadi88/one-man-startup/.cursor/rules/skill-handoff.mdc

one-man-startup / rules

Cursor rule16 starsChanged 2 days agoMITView on GitHub

The format for passing work between agents - a brief going out, a short report with file paths coming back. Use for every delegation and every hand-back.

We split this skill-handoff.mdc into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

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

    Update docs/work/<id>/status.md with the step and its state.

    line 37
  2. Very specific:

    When reviewer, security, legal or a draft the founder pasted in comes back with findings:

    line 64
  3. Distinctive:

    Read all of them before answering any.

    line 66
  4. Distinctive:

    Verify each against the code before agreeing: open the file, run the test.

    line 68

≈ “Common” is an estimate: at least 1 in 5 of the 6,350 Cursor rule 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
skill-handoff.mdc
---
description: "The format for passing work between agents - a brief going out, a short report with file paths coming back. Use for every delegation and every hand-back."
alwaysApply: false
---

<!-- Generated by tools/sync_adapters.py from skills/handoff/SKILL.md. Edit the source, not this file. -->

# Hand-off

## The brief (orchestrator to agent)

- **Objective:** one sentence, with the requirement ids.
- **Scope:** the requirement areas the change touches, set by the orchestrator; `qa`'s regression
  run covers every row in them (TPL-103).
- **Read:** the exact files, nothing else unless stated.
- **To a reviewer:** the artifact, its rows and its contract, never the author's hand-back
  summary: the review starts from the work, not the story about it. A ruling to review is
  named by its number in `status.md`, never quoted, and the reviewer reads it there.
- **Produce:** the artifact and its path.
- **Boundaries:** what not to touch, what needs the founder.
- **Done when:** the checks that must pass, the gate this feeds.

## The hand-back (agent to orchestrator)

- **Status:** done, blocked, or done with gaps.
- **Where:** the folder, branch and commit the work is in, from `git rev-parse --show-toplevel`,
  `git branch --show-current` and `git rev-parse --short HEAD` run at hand-back, and for anything
  running (a preview, the live site) which version it is serving. The orchestrator compares them
  with its own; a hand-back whose folder, branch or running version differs is not done (TPL-115).
- **Artifacts:** paths only.
- **Summary:** three lines at most.
- **Verified:** what was run and what it said.
- **Not verified:** what could not be checked, and why.
- **Opened:** the files read, so nobody reads them again.
- **Needs the founder:** decisions surfaced, each with a recommendation.

Update `docs/work/<id>/status.md` with the step and its state. It opens with four lines,
`Track:`, `Now:`, `Next:` and `Waiting on:` (the founder, an agent's name, or nothing), then
`## Where we left off`: a few sentences, prose only, that let anyone resume after days away or a
cleared context (done, next, what not to redo; TPL-127, after Cline's memory bank,
https://docs.cline.bot/features/memory-bank, read 2026-10-30). Rulings
go under `## Rulings`, one `- ` line each; the founder's status page reads them
(`python tools/status_report.py`).

## Agents working in parallel

Nobody edits a generated or shared file (the index in `docs/REQUIREMENTS.md`, `docs/TEAM.md`,
another area's rows) while others are working. Each writes its own files; the orchestrator runs
`python tools/checks/req_check.py --fix` and settles shared rows once, at the end.

## Rulings

An agent that meets an ambiguity the spec does not settle, and resolves it rather than stopping
(a name, a default, an edge case the rows do not reach), writes a ruling in
`docs/work/<id>/status.md` as it goes: **what was decided, why, and what it costs if wrong.**
Anything on the founder's list (targets, thresholds, prices, scope, legal, permissions, outward
steps) is never ruled; it stops and goes up. A ruling that crosses a module boundary or cannot
be undone cheaply goes to `reviewer` before the build continues. The hand-back lists every ruling, and the
orchestrator puts them in front of the founder at the next stopping point so any can be undone
while it is cheap. A ruling nobody wrote down is an assumption nobody can find.

## Answering a review

When `reviewer`, `security`, `legal` or a draft the founder pasted in comes back with findings:

- Read all of them before answering any. If one is unclear, ask about it before implementing
  the others; related findings are misread when only half understood.
- Verify each against the code before agreeing: open the file, run the test. A finding can be
  wrong about this codebase, break something the reviewer did not see, or add a feature nothing
  uses (grep for callers; if none, ask whether to remove instead).
- Agree by restating the fix in one line, or by doing it. Never "you're absolutely right";
  never announce an implementation before verifying. Disagree with evidence (the test, the
  line, the decision it conflicts with), and put a disagreement about architecture or scope to
  the founder through the orchestrator.
- Implement one finding at a time, each with its test: blocking ones first, then simple, then
  complex. Re-run the sweep after each.
- A re-review reads only what changed. For documents not yet committed, copy the reviewed
  files aside before fixing and give the reviewer `diff -u` of the copy against the fix. Never
  commit with the hooks off to make a diff.
- Reply to each finding where it was made (`review.md`, or the pull request thread) with what
  changed, or why not.

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.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.