Skip to content
Back to skills

Done

ASecurity

Use when claiming a change done, asking for acceptance or a merge, inviting someone to try a build (a bug bash, dogfood run, or visual or design review), requesting a specialist review, or landing reviewed work.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 16, 2026
ai-agentsbashgit

Security analysis

A100/100

Pro scans all 4 files and shows the line behind each finding

Scanned October 16, 2026

npx -y skills add alleneubank/agent-profile --skill done --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Done?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Done
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/alleneubank-done/badge)](https://www.skillsdirectory.com/skills/alleneubank-done)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: done
description: Use when claiming a change done, asking for acceptance or a merge, inviting someone to try a build (a bug bash, dogfood run, or visual or design review), requesting a specialist review, or landing reviewed work.
---

# Done

Done means merged with an evidence record. Acceptance scenarios, written before
the code as things a user or calling agent does and sees, set how much
verification is enough: drive them, plus a high-risk change's own risk
scenarios, and nothing beyond them. The human approves scenarios for features;
the author writes them for fixes. Copy-only changes, and logic whose tests show
what its callers get, need no scenarios beyond those tests. Work verified but
not yet landed (local-only, or awaiting a grant) is reported as verified, not
done.

## Evidence record

Keep it in the pull request body's Evidence section (`git-best-practices`),
or the task record when there is no pull request:

- **Head and build:** the commit and where it ran (localnet, testnet, vendor
  sandbox, simulator).
- **Scenarios:** one row each with surface, what was observed, and a capture
  for UI. A scenario with no observation is not done.
- **Changed behavior:** what a user does differently, per changed control,
  with production's behavior beside it.
- **Verdicts:** each invited role's verdict file and the head it saw.
- **Noticed:** anything you saw that departs from the intent, even where its
  scenario passed. Passing a scenario does not clear what you noticed.
- **Not checked:** what was skipped and why. Before calling a guarantee
  held, list the input classes it covers and the check that drives each; a
  class no check drives reads "not checked". A red negative control shows
  only that the check catches the break it names.
- **Author run:** the surface, the head, and that the author drove each changed
  flow. A guest's run is a separate row.
- **High-risk:** the plan approval, and the rehearsal record (environment, head,
  scenarios, gaps). If there is no rehearsal, the draft reason.

## Invite others to check your work

Give each role its charter by path, plus the inputs its charter lists, and
nothing else: not your reasoning, not your diff unless the charter asks for
it. Every review brief also names the neighbors your checks do not drive: nearby
input classes, callers, and paths. Each returns a verdict file.

| Role | Invite when | Charter |
| --- | --- | --- |
| Guest | whenever a build is worth trying; at the merging head for a new feature or changed flow | [guest](references/guest.md) |
| Design reviewer | a change touches UI | [design reviewer](references/design-reviewer.md) |
| Specialist | a change is high-risk (`AGENTS.md`) | [specialist](references/specialist.md) |

A guest shares your host and uses the build you are already running. Point it
at the build, not at a setup of its own, and say whether that environment is
disposable or live and shared, with its funding budget when it is live. Run the specialist through a
read-only, profile-free launcher, one reviewer per round unless the human asks
for a second. Fix what a verdict finds, then invite a fresh guest or reviewer
at the new head for what the fix could change; the one that saw the last
round tends to confirm its own finding rather than look around it.

## Land

Whoever lands (you under a grant, or a coordinator) checks, in order, and
stops at the first gap:

1. Every scenario has an observation at the head that would merge, including
   a high-risk change's own risk scenarios. Earlier evidence counts only
   while what it covered is unchanged.
2. The author drove each changed flow end to end on the closest-to-live
   surface, and the evidence record's author-run row names that surface and
   head. A guest's run does not fill this row.
3. A new feature or changed flow has a guest verdict with every scenario
   passed, current for that head: a fix after the verdict needs a new guest
   run only for the scenarios it could change. A human-only step (a physical
   device, biometrics) the guest marked blocked passes once the human records
   its result.
4. A UI change has a design verdict at that head with no open blocking
   finding.
5. A high-risk change has a recorded plan approval, a specialist verdict at
   that head with no open blocking finding (or the human's recorded waiver),
   and an author rehearsal at that head in the closest real environment,
   recording environment, head, scenarios, and gaps. Without the rehearsal
   it stays a draft, saying why. A code-law waiver does not cover these gates.
6. Required checks are green, not skipped or missing.

A gap sends the work back, naming the missing item. When every item passes,
land under a grant naming this repository and branch. Otherwise ask the human,
linking the evidence record and stating the review status: who reviewed, at
which head, or that no review ran. A yes given knowing no review ran is
recorded as the human's waiver.

## Red flags

- "Done" with a scenario that has no observation, or a behavior change with
  no scenarios
- A verdict from an earlier head presented as current after a change it covers
- A guest given the diff or the author's reasoning
- A merge ask for a high-risk change that does not state its review status

Files in this skill

  • SKILL.md5.2 KB
  • references/design-reviewer.md1.6 KB
  • references/guest.md4.4 KB
  • references/specialist.md1.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.