Skip to content
Agentmart
Submit a tool

customer-delivery

Enforce delivery from a verbatim requirement through pre-work knowledge selection, production evidence, and disposition; excludes enabling-only closure.

Skill by Arcanada-one

IndexedMIT★ 16
Write a review

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/arcanada-one-customer-delivery

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

Enforce delivery from a verbatim requirement through pre-work knowledge selection, production evidence, and disposition; excludes enabling-only closure.

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 Oct 25, 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/customer-delivery

What's inside

Skill name
customer-delivery
Description
Enforce delivery from a verbatim requirement through pre-work knowledge selection, production evidence, and disposition; excludes enabling-only closure.
License
MIT
Instructions
185 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Customer Delivery

A user-facing requirement is delivered only when a verbatim customer statement
flows through pinned knowledge, an implementation delta, red/green evidence, a
merged revision, a deployed revision, live production evidence, and an explicit
operator visual disposition — in that order, with no missing edge.

Enabling artifacts — tools, documentation, tests, CI runs, ledger entries,
receipts — support closure but never substitute for it.

## U1. Verbatim atomic requirements

- Store every customer remark verbatim; a paraphrase is metadata, never the
  source of truth.
- Decompose each remark into atomic, stable Requirement IDs.
- Corrections are append-only: record the superseding statement and retain the
  superseded one.

**Fail** if a requirement cites a paraphrase, an aggregate summary, or an
unstable ID as its origin.

## U2. Acceptance tuple

Every Requirement ID carries:

- originating source and exact quotation;
- affected product and surface or route (surface class);
- locale, viewport, and theme dimensions when applicable;
- observable before-state and required after-state;
- evidence method and responsible owner;
- applicable role, skill, blueprint, constraint, policy, and success criterion;
- delivery task, code/content paths, and target environment;
- customer disposition: pending, accepted, rejected, or superseded;
- visibility classification: enabling or visitor-visible;
- receipt pointer to its coverage receipt.

Bind expectations through schema v4 by declaring `customer_derived` on every
wish. A `true` declaration carries the Requirement ID, surface class,
visibility, and receipt pointer; a `false` declaration carries none of those
four fields.

At least one acceptance criterion must assert an observable visitor-facing
change on the live production product. A rendered test-environment comparison
is mandatory supporting evidence but cannot replace the live result.

**Fail** if a user-facing requirement is satisfied only by knowledge,
documentation, test, CI, or ledger output.

## U3. Pre-work knowledge pinning

- The applicable knowledge is selected before implementation starts. Before
  `implementation_started_at`, the issued contract pins current
  admissible revisions of the applicable role, skill, blueprint, constraint,
  policy, and success criterion — each with an immutable revision, digest, and
  timestamp.
- `Gap` or `Unbound` may authorize research or capability creation but may
  never be represented as blueprint-applied product delivery.
- Selection after implementation start is post-hoc attribution and fails.
- Carry the selected bindings and task-relevant evidence into delegated or
  isolated stages through approved readable references or a sanitized bundle;
  a parent-only conversation or inaccessible host path is not inherited context.
  Preserve the original selection time, revision and digest, and distinguish
  evidence measured on a named environment/revision at its observation time
  from inferred claims. Verify availability before affected work; missing
  material remains an explicit gap, not permission to reconstruct protected
  originals or invent a prior selection.
- Inherited measurements are historical evidence, not fresh verification of
  the current target. Preserve original `verified:` / `assumed:` provenance
  metadata and apply the replay downgrade and revalidation rules in
  `${DATARIM_RUNTIME:?}/skills/session-handoff-writer/SKILL.md` and
  `${DATARIM_RUNTIME:?}/skills/session-handoff-replay/SKILL.md`; a copied receipt,
  matching document digest, or rewritten timestamp cannot refresh a live check.
- For classified sensitive sources, use the existing constraints/policies selection to pin the task's metadata-only sanitized binding, following `${DATARIM_RUNTIME:?}/skills/security/SKILL.md` § Sensitive source context boundary. Record its revision, digest and actual selection time before future affected reads or work. A classification discovered during work must be preserved prospectively; this does not establish retroactive pre-work compliance or clear an earlier missing-selection failure.

**Fail** if any required knowledge kind is missing, `Gap`/`Unbound` stands in
for a delivery-bound contract, or a revision is selected after work began.

## U4. Coverage receipt

Emit a machine-readable acyclic mapping per delivery:

`Requirement -> selected knowledge revisions -> implementation delta ->
red/green evidence -> merged revision -> deployed revision -> live evidence ->
customer disposition`

- A missing edge means `NOT MET`.
- Receipts are deterministic and mutation tested: removal of any required edge
  must turn the gate red.

**Fail** on a cycle, a missing edge, a mutable revision reference, or a receipt
that cannot fail when an edge is removed.

## U5. Closure semantics

A user-facing task or epic cannot close while:

- any originating customer review is open or changes-requested;
- any required live evidence is absent;
- production is behind the accepted revision; or
- operator visual acceptance is pending.

Executable rules, not guidance:

- `green CI != deployed`
- `deployed != visually accepted`
- `enabling work != customer outcome`

Epic state is derived from the Requirement graph and child acceptance states,
never from an independently authored prose status.

**Fail** if a task closes on CI, merge, or internal review alone.

## U6. Review-to-evolution

Classify every review finding exactly one of:

`ABSENT`, `WEAK`, `STALE`, `MIS_SCOPED`, `NOT_BOUND`, `NO_CANON_CHANGE`

- The first five require an artifact revision or new artifact plus a

From Arcanada-one/datarim, MIT.

More from Arcanada-one/datarim

Signals

Stars
16
Forks
1
Open issues
0
Last commit (this folder)
Oct 4, 2026 (24 days ago)
Last release
Oct 22, 2026 (5 days ago)

From GitHub (fetched Oct 24, 2026).

Details

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

More from Arcanada-one

Cite this page

customer-delivery skill. Agent Mart. https://agentmart.dev/skills/arcanada-one-customer-delivery (last updated 2026-10-24)