agentleFS
nateprichnateprich/ai-workbench/coding-implementation/SKILL.md

coding-implementation

Skill1 starsChanged 7 days agoMITView on GitHub

Deliver software from a raw idea, a thinking handoff, an agreed spec, or a bounded coding work item. Clarify consequential choices, define acceptance, coordinate implementation, and independently verify real behavior. Use for coding projects and software changes; keep planning-only requests within planning and leave non-coding production to suitable capabilities.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read a stage reference before doing that stage, including when entering directly as a builder or verifier.

    line 21
  2. Distinctive:

    Use only the access and external actions already authorized.

    line 31

≈ “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

Coding Implementation

Carry a finite software task from its current state to the agreed, verified result. Use one preparation boundary: Needs definition → Ready → Building → Verifying → Done. Shaping, specification, and splitting happen within preparation; there is no separate Shaped checkpoint or approval between them. No upstream framework, tracker service, worktree system, or companion skill is required.

Start at the right stage

Establish what is settled, what action the user authorized, and whether you own the project or one delegated item. Reuse accessible context and project instructions, including saved tracker setup. Do not restart discovery for an adequate spec, repeat settled questions, or rediscover setup for each item.

Current needRead and apply
Durable tracking is needed and configuration is missing or concretely staleProject setup, using the GitHub or local template
Consequential behavior, acceptance, coverage, or work boundaries remain undefinedPrepare the work
Work needs selection, implementation, integration, state transitions, or resumptionBuild and coordinate
An item or integrated result is ready for independent checks; findings need adjudication or follow-upVerify and close

Read a stage reference before doing that stage, including when entering directly as a builder or verifier. A coordinator also reads verification guidance before commissioning a verifier. Follow the same entry rules for a direct request and a handoff.

When material thinking remains, use Thought Partner if available and suitable, honoring an explicit skill choice. Discover capabilities through the current host, load the chosen instructions, and return the thinking result to this workflow. If it is unavailable, do the necessary clarification here. Do not require installation or assume a skill name establishes its availability or ownership.

Preserve authority and boundaries

  • Carry the authorized action, scope, limits, and pending consequential decisions across every handoff. Preserve the relevant user instruction or a concise accessible reference. An agreed direction or permission to draft a spec does not by itself authorize building. Earlier execution authorization remains valid; do not add a new approval ritual.
  • The authoritative spec is the completion contract; item briefs cannot silently override it. Make consequential tradeoffs visible before treating them as settled. Inspect code and facts when they can resolve uncertainty.
  • Routine implementation choices, repairs to meet the spec, and verification within an item remain delegated. A spec redesign or implementation redesign beyond that item's bounds returns to the user with evidence, the smallest viable options, and a recommendation. Continue unaffected authorized work.
  • Additional abstractions, infrastructure, dependencies, and refactors need an agreed requirement, a relevant regression, or a concrete consequential failure under intended use. Future scale, style preferences, and “best practice” alone do not expand scope. Leave unrelated cleanup and possible future improvements out of the active work; do not create a backlog or improve the workflow unless requested.
  • Use only the access and external actions already authorized. Implementation and testing do not grant permission to install, publish, deploy, send, or change live systems beyond that scope.

Own the finite result

Once preparation and execution authority are sufficient, the coordinator selects eligible items, manages dependencies and findings, integrates accepted results, and independently verifies the assembled user path. The coordinator owns state transitions and acceptance; builders return bounded results and evidence. Use fresh builders and separate fresh verifiers when supported and authorized.

Use a short spec and one work record for a small change. Larger work needs coverage of every agreed outcome, while later items may remain Needs definition. Parent progress is derived from that coverage and accepted results, not merely the state of existing tickets. Do not introduce portfolio scheduling, autonomous intake, dashboards, a new runtime, or automatic workflow improvement.

Completion requires accepted coverage of the spec and required checks on the current artifact and applicable contract, with no substantiated serious concerns. Builder tests, empty reviews, mocks, and structural checks cannot establish an unobserved required effect. Missing required evidence stays outstanding. Follow verification and closure, report evidence and limitations, then stop when the finite work is complete.

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.