Skip to content
Agentmart
Submit a tool

deliver

Deliver an accepted spec autonomously, from failing acceptance tests to a pull request with green checks.

Skill by taecontrol

IndexedMIT★ 1
Write a review

Chalk Labs

Need help deploying deliver?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/taecontrol-deliver

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Deliver an accepted spec autonomously, from failing acceptance tests to a pull request with green checks. Use when a spec from `shape` is accepted, or for a bug fix whose reproduction and expected behavior are agreed.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Nov 1, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/deliver

What's inside

Skill name
deliver
Description
Deliver an accepted spec autonomously, from failing acceptance tests to a pull request with green checks. Use when a spec from `shape` is accepted, or for a bug fix whose reproduction and expected behavior are agreed.
License
MIT
Instructions
37 lines in SKILL.md · view on GitHub

Files (2)

  • SKILL.md
  • openai.yaml · agents/

SKILL.md (first lines)

# Deliver

Take one accepted spec to a pull request the human only has to merge. The spec's examples define success; the acceptance tests make them executable before any production code exists; the rest of the work makes them pass and leaves the code in good shape. Run to the end on your own.

## Start

Read the spec at the path you were given, every source its Context and Decisions point to, the project's coding standards, and its gate configuration. Require `Status: Accepted`. For a bug fix, the agreed reproduction and expected behavior are the spec. Work on the current branch in the current worktree.

## 1. Acceptance tests

Turn each `test` example into an executable test at the spec's test boundary, with the project's verification method, under the spec's real target conditions. Name each test for the behavior it proves, in words a reader without the spec follows, without the example's ID. Run them and watch each go **red** for the right reason: the behavior is missing, not the setup broken. A test for something that must never happen must be able to fail. Commit the tests on their own; when the project's hooks reject failing tests, mark them as expected failures until the code makes them pass.

These assertions are the contract. Glue around them, such as setup, selectors, and helpers, may change as the code takes shape. An assertion changes only when it misstates its example; list each such change for the [report](#report).

## 2. Make it work

Make the tests pass, then capture each `capture` example from the running product. Commit as you go. Done when every acceptance test is green, the project's gates pass, and the evidence is captured.

What must outlive the spec goes into the change: behavior is already in tests; rationale goes into an ADR through `adr`; project-wide rules are drafted where `agents-md` keeps them; and a new or changed domain term goes into the project's glossary, in its existing format, when the project keeps one. The human accepts all of it with the pull request; a project without a glossary gets one only when the human agrees. What only a later issue needs goes into that issue.

## 3. Refine

Run `refine` in a fresh context, such as a subagent, with the spec, the diff range from where delivery began, and the assertion changes. When the harness offers no fresh context, run it yourself and say so in the report.

## 4. Ship

Run `ship` once `refine` finishes. When `refine` stopped on blockers, stop with its report instead. A production change made after `refine` finishes, such as a fix for a CI check or one the human asks for, gets a fresh review of that change and its callers before the pull request is ready.

## Decisions along the way

Decide reversible choices yourself and record the ones the human would want to know about for the report. A one-way door the spec does not settle, such as a migration, a new contract that writes, or a change to a seam an accepted ADR governs, is not one of them. An accepted program design constrains them: follow it where the code agrees, and name each departure from it in the report with its reason. Stop and ask when the change needs one, or when evidence shows that an example, a decision in the spec, or an accepted one-way door is wrong, and continue with work that does not depend on it. The examples are the contract, but a number that sets how much to verify, such as a repetition count, a sample size, or a margin, is a decision like any other: before a verification that runs long or repeats, time one run, and when the whole would cost out of proportion to what it adds, that is evidence its decision is wrong. Put anything to the human, a question or a stop such as `refine`'s blockers, so they can answer it without opening the spec: the behavior at stake in plain words, the options, and your recommendation, written with `writing-for-people`.

## Report

End with a report to the human, in their language, in your final message rather than the pull request, written with `writing-for-people`. It carries what the pull request leaves out because only someone who knows the spec can use it: the pull request and its check state, how each example is proven, every deviation from the spec and from the program design with its reason, each changed acceptance assertion, the decisions you made on your own, and `refine`'s report.

## Done

The draft pull request is open, its checks are green, every example is proven by a test or a capture, and every deviation from the spec is named in the report. The human marks it ready for review and merges.

From taecontrol/skills, MIT.

More from taecontrol/skills

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Oct 23, 2026 (12 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from taecontrol

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get deliver running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying deliver?

CallGet help
Cite this page

deliver skill. Agent Mart. https://agentmart.dev/skills/taecontrol-deliver (last updated 2026-10-24)