AgentStack
SKILL verified MIT Self-run

Deliver

skill-mblauberg-provenant-deliver · by mblauberg

Use for taking an approved research, analysis, document, or agent-product outcome through evidence, review, and user acceptance. Not for software-only work, unsettled scope, or release; use implement, scope, or release.

No reviews yet
0 installs
0 views
view→install

Install

$ agentstack add skill-mblauberg-provenant-deliver

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Deliver? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Deliver

Run the domain-neutral lifecycle kernel. This skill coordinates existing capabilities; domain skills still own domain methods.

Entry

Require an approved intent, acceptance criteria, minimum risk tier and bounded authority. Consequential scope/design, disclosure, one-way doors and risk downgrades need explicit user approval. If intent is unsettled, use scope.

Select one base profile from config/delivery-profiles.json: software, research, analysis, document or agent-product. Add the high-stakes overlay when source authority, privacy or qualified review matters. Projects may strengthen a profile, never weaken kernel gates silently.

Lifecycle

  1. Create .agent-run//RUN.json from templates/RUN.template.json and

bind intent, design and authority by digest. When the delivery belongs to a coordinated project, bind the top-level fabric_relationships to its project session, coordination run and workstream/lead IDs per [the receipt contract](references/contract.md). Use the explicit independent not_applicable form rather than inventing parents.

  1. Record each state transition. No state may jump an approval, evidence,

review, acceptance or release gate.

  1. Execute through the relevant skills. Software routes execution through

implement; stochastic behaviour routes through evaluate; failures use diagnose; substantial parallel work may use orchestrate.

  1. Produce profile-required deterministic evidence before judgement evidence.

Every gate links to a typed artifact or receipt. At acceptance, a stochastic gate must bind and hash-verify a passing canonical evaluation-run receipt; copied scores or sampling metadata are not evidence. Retain failed or incomplete evaluation receipts as non-gating history.

  1. Review independently with lenses selected from the dependency cone.

Substantial+ follows HARNESS.md: targeted lenses plus the other primary; distinct-family review is used when available, with terminal pressure made stronger and skipped optional legs recorded.

  1. Repair at most twice. Scope/design drift returns to the user gate.
  2. Validate from the project root with

"${AGENTS_HOME:-$HOME/.agents}/skills/deliver/scripts/validate_delivery.py" .agent-run//RUN.json --workspace-root "$PWD" --verify-hashes (plus a digest-bound --project-policy when used). awaiting_acceptance is machine-ready, not complete.

  1. User acceptance and external release are separate. Define observation

before release; close only after its evidence window passes. Feed incidents and recurrence into retrospect and the next scoped cycle.

Boundaries

Delegates may only narrow authority. One writer owns each shared source surface. Artifact manifests classify canonical, evidence, handoff, scratch or external; cleanup removes only expired, run-owned scratch with explicit authority. Filesystem receipts remain truth when Herdr or another transport is unavailable.

implement uses this same receipt with profile software; no parallel implementation receipt format exists. Live task or membership projections may link to it but never replace its canonical acceptance evidence.

For audience-ready HTML, apply the [interactive-document boundary](references/interactive-documents.md).

Adapter-absent path

Console, Herdr and GitHub are optional. Continue from canonical project artifacts and emit the skill-owned artifact kind in [portable-workflow.v1.json](portable-workflow.v1.json). That filesystem artifact records workflow evidence; it never grants acceptance or release authority.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.