Skip to main content
skill CX Codex

deliver — Claude Code skill: workflow and install

This Claude Code skill carries a released charter through to verified completion. It starts with /deliver or when an agent assigns it the executor role.

Apache-2.0 ★ 869 updated 2026-11-16

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add sno-ai/sno-station-skills --skill deliver -a claude-code
Open on GitHub

What deliver does

The skill reads the charter and existing work, then plans dependent steps and carries them out with the project's own tools. After each step, it updates the Done and Next lists with sno handoff-checkpoint when that command is installed. Each success check is backed by a real command run or inspected evidence, recorded through sno deliver-proof.

Once the work is done, the skill gets one independent review, fixes in-scope defects, and runs sno deliver-proof check. It writes the report and sets delivered only after that check passes. sno deliver-proof requires Bash 4+; on macOS without GNU coreutils, Python 3 is also needed. Its self-test needs Python 3 as well. If the command is unavailable, run scripts/deliver-proof from the skill's directory.

What it helps with

  • carry out work defined by a released charter
  • verify each success check with a real run
  • hand unfinished work to the next executor

Example prompts for the agent

  • Complete the release-fix charter against its success checks.
  • Continue the checkout charter from its Next list.
  • Review the docs-refresh result and write the closing report.

How to install deliver

1Skills CLI (Vercel)

npx skills add sno-ai/sno-station-skills --skill deliver -a claude-code

2Manually: put the files in place

~/.claude/skills/skill/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/skill/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about deliver

How do I install deliver in Claude Code?

Run npx skills add sno-ai/sno-station-skills --skill deliver -a claude-code. The skill file is available at ~/.claude/skills/skill/SKILL.md.

What does deliver do?

It carries out a released charter, records real evidence for each success check, gets one independent review, and closes the work with a report.

What does deliver need to run?

sno deliver-proof requires Bash 4+ and, on macOS without GNU coreutils, Python 3. Its self-test also needs Python 3. If the command is unavailable, run scripts/deliver-proof from the skill directory.

Source text

SKILL.md · Apache-2.0
---
name: deliver
description: "Carry out a released charter to a proven result: plan only as much as the work needs, do the work through the project's own workers, prove every success check with a recorded real run, get one independent review, and close honestly. Invoke explicitly as `$deliver <charter>` (Codex) or `/deliver <charter>` (Claude, Cursor), or the way your agent invokes a skill, or when a dispatch names you the executor of a charter."
requires:
  programs: []
  harness:
    - {slot: 4.shell, need: required}
    - {slot: 4.file-read-write, need: required}
---

# deliver

Carry out the named charter. Its `## Success checks` are the definition of done; nothing else
is. Read the whole charter and the existing work first. The owner (whoever owns the work,
often the user) can override this skill and any plan an agent wrote. Any kind of project uses
it: for code and tests also read `references/proof.md`.

Prerequisites: `sno deliver-proof` needs bash 4+ and prints one message and stops when it is
missing; without GNU coreutils (macOS) it also needs python3, and its selftest needs python3. On
macOS, install bash 4+ (for example with Homebrew) and put it first on PATH. If `sno deliver-proof` is not
available, run `scripts/deliver-proof` from this skill's folder instead.

## Keep the work useful

- Before adding a prerequisite, test, record or approval step, name the requested result it
  proves and what changes if it fails. If nothing changes, leave it out.
- Check an uncertain fact by reading or calling what exists. If checking needs something built
  first, building it is the first task, not a permission gate.
- Records, telemetry, formatting, identifiers, checksums and review sign-offs never block the
  work. Log the failed action, its place and the actual error, and continue. Do not turn such
  a failure into an assertion, a silent return or a swallowed error. If the main operation
  fails, say so with its cause; never report success. Stop only what cannot proceed and keep
  independent work moving.
- Time is finite and proof shares it: estimate a run from existing timings before starting,
  test only what changed and what directly depends on it, and by default run a full suite only
  when the owner explicitly asks. Do not stop a useful run just because an estimate has passed.
- By default add no security check or new gate the owner did not ask for; propose it as a
  question that says what it would check and what it would block.

## Steps

1. **Start.** If the charter is missing, is not `released`, or has a success check nobody could
   observe, say so once and have `charter` revise it with the owner. Never invent a decision.
   If `<charter-name>.state.md` sits beside the charter, an earlier executor stopped partway:
   run `sno handoff-checkpoint --verify` on it (the `handoff` skill ships the command), read its Done
   and Next lists, treat Done as finished, and continue from Next instead of starting over.
2. **Plan, only as much as needed.** Work with dependent steps gets a `## Plan` in the charter:
   the steps in order and who does each. One or two steps need none. A plan of more than three
   steps gets one independent review, then work starts; there is no second plan document. A
   product decision the charter does not settle is recorded in the plan and asked once, while
   independent work continues.
3. **Do.** Use the project's own workers and tools: for code, its coder, test and cleanup
   skills when installed; for documents, its writing and review tools. Build the smallest real
   integration early when later steps depend on it. Keep existing behavior unless the charter
   says otherwise. Reproduce a defect before fixing it.
   After every finished step run `sno handoff-checkpoint <charter-name>.state.md` (beside the
   charter) and update its Done and Next lists, so another agent can take over at any moment.
   If the command is not installed, skip the record and keep working.
4. **Prove.** For each success check, record one of:
   - `sno deliver-proof run <charter> <n> -- <command>`: the real command, run in the current
     directory; its output and exit status are kept as the log and the table row.
   - `sno deliver-proof see <charter> <n> <evidence-file> "<what was observed>"`: when inspection is
     the proof (a produced document, a message at its receiver, a screen). The evidence file is
     a nonempty file, named relative to the current directory.

   Prove the result where the requester would see it: read back the saved value, open the
   produced file, read the message at its receiver. Never substitute a mock's answer, a
   successful exit, a call count or the program's own narration. The expected value comes from
   the charter or independent data. Never edit the Proof table by hand, never mark a failed or
   unrun check passed, and after a repair rerun only the affected checks.
5. **Review once.** One independent review of what changed (`peer-review` when installed,
   otherwise a second agent given the charter and the changed work). Fix ordinary-path defects
   in scope; report the rest without growing the job.
6. **Close.** Run `sno deliver-proof check <charter>` and read its output. Only when it exits 0,
   write `## Report` (what worked, which checks ran and how, anything unfinished) and set
   `status: delivered`. A failing record step is logged and repaired on its own; it does not
   undo working results. Tell whoever dispatched you the result and what remains.

Publish, send to other people, spend, delete, deploy and commit only within the owner's
authorization. If `~/.config/sno/decision-rights.md` exists (the `pl` skill creates it), it
lists what needs the owner. Do not absorb the next charter's work. When you are blocked on a
missing input, state the assumption you proceed under in the plan and keep working.

© sno-ai · Apache-2.0 · sno-ai/sno-station-skills

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set40

The whole set →
cos-review The cos-review Claude Code skill audits decisions, completed work, and estimates after a declared night shift. COS core routes it in when the owner returns or a batch of closes lands. CX Codex +1 ★ 869repo stars cos-evolve The cos-evolve Claude Code skill turns recurring management failures into rules for COS or PL. COS loads it after the owner activates the role and routes a repeated failure to it. CX Codex +1 ★ 869repo stars codex-coder The codex-coder Claude Code skill assigns coding work to a Shell, TypeScript, Python, or Rust specialist. It activates for Codex coding requests, language migrations, and tasks that change files in several languages. CX Codex +1 ★ 869repo stars codex-coder The codex-coder Claude Code skill hands coding work to Codex through codex exec. It routes Shell, TypeScript, Python, and Rust tasks, including language migrations. CX Codex +1 ★ 869repo stars tag-team The tag-team Claude Code skill assigns one task to two agents: one implements it, while the other reviews the changes. It activates when asked to pair agents for implementation and review. CX Codex +1 ★ 869repo stars subscription-quota-check subscription-quota-check is a Claude Code and Codex skill that reads remaining subscription quota without using it. Run it on request or after a provider rejects work for quota reasons. CX Codex +1 ★ 869repo stars rotate-agent The rotate-agent Claude Code skill switches to another model or agent when the sender’s quota reaches a threshold. Once armed, it checks that quota and tries candidates in the configured order. CX Codex +1 ★ 869repo stars reach This Claude Code skill connects agents through Reach. Instant Messenger waits for a reply; Agent Email stores messages and tasks for later. CX Codex +1 ★ 869repo stars

Project sponsors

Companies that keep this analytics open