agentleFS
nesnilnehcnesnilnehc/ai-cortex/skills/execute-delivery-workflow/SKILL.md

execute-delivery-workflow

Skill8 starsChanged 2 months agoMITView on GitHub

Run an existing project-owned artifact delivery workflow for a specified source and destination, verifying the resulting artifact at the target. Use for one build-and-deliver operation; missing or broken workflow design belongs to establish-delivery-workflow.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the workflow's project-owned handoff: invocation/parameters, supported targets, source-state policy, environment, verification gates, …

    line 18
  2. Very specific:

    Treat implicit tracked-source/configuration/version/description edits as excess effects too.

    line 32
  3. Very specific:

    If the workflow is missing, incompatible, or incomplete, report the exact gap and hand it to establish-delivery-workflow when the user …

    line 35
  4. Distinctive:

    When promotion requires the same artifact, preserve its bytes and embedded version.

    line 45
  5. Distinctive:

    After a timeout or uncertain upload outcome, inspect the destination before retrying.

    line 58
  6. Distinctive:

    Do not automatically reset/stash unrelated work, roll back external state, remove versions, or change credentials.

    line 61

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Execute Delivery Workflow

Complete one authorized artifact delivery through the project's authoritative workflow. Delivery means required verification passed and the exact artifact is confirmed at its destination. A successful CI job, local build, or upload response alone is insufficient. Adapt to the project's environment and artifact format rather than prescribing universal commands.

Inputs and execution contract

Require the project and component, source reference/state, exact version or project-defined version source, target destination/channel, and the applicable workflow entry point. Discover these from explicit instructions and current project configuration; honor established defaults only when their meaning is unambiguous. Do not invent a release destination, requirement scope, or version policy. A request to deliver to a named target authorizes that delivery within its stated scope; do not seek repeated approval for each stage. A request limited to a build, preview, or dry run does not authorize upload.

Target version and delivery scope may be decided offline or in existing project records. Ask only for a specific missing or conflicting input; do not require a version-planning skill. Change-description preparation is separate: consume its result when project/target rules require it. Ordinary branch/test builds need no universal release-note gate; candidate/formal releases follow the selected project's rules.

Read the workflow's project-owned handoff: invocation/parameters, supported targets, source-state policy, environment, verification gates, expected artifact, credential references, destination check, and recovery behavior. These may be implemented by scripts and CI configuration with linked documentation; no particular contract file or CI vendor is required.

Inspect current branch, HEAD, working-tree/staged changes, and the actual component before execution. Bind the run to an exact source state. Do not silently discard local work or deliver uncommitted changes under a commit-only identity. Follow the project's policy for dirty checkouts and record a sufficient identity for any explicitly supported working-tree build; if it cannot be represented reliably, report that gap.

Resolve the selected source to its exact commit SHA, including for main, feature or defect branches; a tag is not mandatory. Compare tags, version files, build parameters and package versions according to the project's version rules. Report conflicts rather than choosing one silently.

By default, execution does not edit tracked source, project configuration, version files or change descriptions. Project-defined outputs, caches and temporary files, plus the authorized artifact upload, are allowed. Project-defined build-time version injection may write into an isolated build directory; retain the source/parameter/final artifact identity and do not label those transformed inputs as a pure commit snapshot. Upstream repairs require their own authorized scope; do not change versions or write descriptions merely to make a delivery check pass.

Check feasibility and scope

  • Confirm required tools/versions and runner availability, dependency setup, expected parameters, destination, and non-secret credential references. Existing read access does not establish write permission.
  • When descriptions are required, check that the existing notes/changelog/list exists and identifies the selected component/version and target source, with a recoverable revision/digest and unresolved issues where relevant. Consume the stated comparison purpose, baseline and source association; do not select a new baseline, reassess requirement membership or rewrite the description. Report gaps to the responsible capability or project mechanism before dependent execution.
  • Check that any reused verification evidence applies to the current source, inputs, environment and payload, whether or not descriptions are required. Missing or stale evidence must be replaced by the applicable project verification before dependent delivery steps; do not imply a previous green result covers changed inputs.
  • Inspect the invoked scripts, lifecycle hooks, and pipeline behavior for effects beyond the request. A release command may also bump, commit, tag, push, deploy, or publish to several targets. Select a documented scoped mode, or resolve the excess scope before running it.
  • Treat implicit tracked-source/configuration/version/description edits as excess effects too. Use an existing suitable mode or stop before effects outside the current authorization; do not modify the source worktree to accommodate build-time injection.
  • Before invoking an entry point that can upload, inspect how it selects, verifies, packages, and uploads the artifact. Establish that the uploaded payload will be the validated payload, including when packaging occurs again. If this cannot be established, use a documented build-only or preview mode if in scope, or stop and report the workflow gap before upload.
  • Prefer a sufficient connected CI/registry/storage tool. Discover operation and status/read-back capabilities; otherwise use the project-supported CLI. A queued CI run is not a completed delivery. Identify the run created for this request and wait for its terminal outcome rather than attaching the result of an unrelated job.
  • If the workflow is missing, incompatible, or incomplete, report the exact gap and hand it to establish-delivery-workflow when the user requests that work. Continue independent read-only assessment, but do not silently redesign infrastructure or fabricate a replacement command sequence.

Execute the established workflow

Invoke the supported entry point with the resolved source, component, version, and destination. Reuse project entry points rather than creating a second implementation of their sequence. Track the actual stages and outputs: build, required verification, packaging, upload, and destination read-back. Stop on failed prerequisites or verification; do not weaken checks or continue to publish an unverified result.

Bind the packaged artifact to the run: record its exact name/version or immutable identifier, source state, runner/build reference, and digest where supported. Check that execution follows the inspected path and the upload consumes the validated artifact. If it deviates or identity cannot be established, stop any remaining upload steps and report the gap and any external effects.

If execution pauses at a project acceptance gate, report awaiting acceptance with the retained artifact identity and run reference. Do not bypass the gate or trigger another build to represent promotion of the tested artifact. A snapshot, prerelease, or stable destination has only the semantics established by the project contract.

When promotion requires the same artifact, preserve its bytes and embedded version; changing its package version or rebuilding produces a different artifact and cannot count as unchanged promotion.

Verify at the destination

Query the actual target used by this run. Check the exact artifact identity and version/build, location, channel pointer if it was changed, and destination digest against the delivered payload when supported. Verify all required outputs for multi-artifact runs. Report per-artifact outcomes; partial success is not complete delivery.

For registries or directories without a comparable digest, use the project's documented alternative identity/read-back evidence and state the limitation. If the documented success criterion cannot be met, report delivery as unconfirmed rather than treating a green job as proof. Presence does not imply production deployment or independent acceptance.

A read-only artifact inspection capability can perform the destination check if available. It must query the artifact source itself; build logs may explain the run but cannot replace that observation. Do not require a particular companion skill to be installed if the necessary read-only operation is otherwise available.

Failure, retries, and resumption

  • On upload failure, retain the validated artifact, digest, source/run identity, and safe retry reference under project retention rules. Do not rebuild by default. If the runner no longer retains the artifact, state that a new build is required and creates a new candidate identity.
  • After a timeout or uncertain upload outcome, inspect the destination before retrying. Avoid publishing a duplicate merely because the client did not receive confirmation.
  • If the target identity already exists, compare content/immutable identity. An identical result can be reported as already delivered when the project allows it; conflicting immutable releases block the run. Mutable snapshot aliases require the explicit policy and distinguishable build identities established by the project; no implicit overwrite or deletion.
  • Reuse earlier verification only while the relevant source, inputs, environment, and payload identity remain valid. Restart only affected steps; a change to the payload invalidates artifact-specific checks. Report retry history instead of implying the first attempt succeeded.
  • Do not automatically reset/stash unrelated work, roll back external state, remove versions, or change credentials. Report partial external effects and the concrete next action if blocked.

Output and completion

Lead with what was delivered and where it can be obtained, or the precise stage that stopped. Include selected source/version/target, actual run reference, verification summary, artifact identity/digest, destination evidence, retained artifacts, and material limitations as needed. Keep secrets out of logs and reports. Use natural prose; expand technical detail only when needed to evaluate the result.

Use delivered, already delivered, awaiting acceptance, partially delivered, failed, or delivery unconfirmed according to evidence. A local-only or dry-run success is build/preview complete, not delivered. Do not claim ready for production from a metadata check. This skill does not select requirements for a release, rewrite project policy, bump/commit/tag/push source, or deploy applications unless those actions are explicitly included in the user's scope and established workflow.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.