agentleFS
emiliodominguezemiliodominguez/flow-skills/skills/flow-specialize/SKILL.md

flow-specialize

Skill1 starsChanged 46 days agoMITView on GitHub

Repository-specific worker and verifier briefs. Use when delegating tasks to agents that need local architecture, lint rules or domain contracts. Feeds `flow-orchestrate` or `flow-work`.

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 target task, repository instructions, configuration, and representative implementation and test files.

    line 14
  2. Very specific:

    Write worker.md and verifier.md to the user's location or .plans/specialists/<topic>/.

    line 20
  3. Distinctive:

    Stop conditions and the plan's correction budget.

    line 29
  4. Distinctive:

    Never preload either role with the desired verdict or tell it to ignore defects.

    line 40
  5. Very specific:

    Default to portable Markdown briefs, usable by an available agent or by flow-work directly.

    line 44
  6. Distinctive:

    Write ownership, acceptance, stop conditions, and evidence outputs are explicit.

    line 60

≈ “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

Specialize a worker

Turn repository evidence into a focused execution brief. Specialization supplies constraints and checks; it does not grant tools, register an agent, or override user scope.

Phase 1: Establish the evidence

  • Read the target task, repository instructions, configuration, and representative implementation and test files.
  • Reuse a current flow-repo-brief, reopening sources for each relevant constraint. Resolve a stale brief or ambiguous task before writing a confident profile.
  • Separate enforced rules from precedent and preference. Copy actual EditorConfig, Prettier, ESLint/Stylelint, type boundary, and documentation rules when present. Do not import DDD, a logging library, a build chain, or function syntax from another repo. Note deliberate local exceptions.

Phase 2: Produce two bounded briefs

Write worker.md and verifier.md to the user's location or .plans/specialists/<topic>/. Keep roles separate and link shared evidence.

The worker brief includes:

  • Purpose, matching task types, explicit exclusions.
  • Repository revision, source citations, relevant files and symbols, input dependencies.
  • Allowed write set and external actions; shared files that need coordination.
  • Task-relevant conventions with provenance and exceptions.
  • Deliverable, behavioral acceptance, exact local commands, evidence format.
  • Stop conditions and the plan's correction budget; report scope conflicts.
  • Output: actual changes and revision, checks and results, open questions, artifact path.

The verifier brief includes:

  • The same unchanged acceptance criteria and downstream use.
  • Independent read access to implementation, tests, and raw outputs.
  • Pragmatic or production profile with relevant runtime and repository gates.
  • A ban on changing deliverables or accepting worker claims without evidence.
  • The flow-verify verdict format, including BLOCKED for missing prerequisites.

Never preload either role with the desired verdict or tell it to ignore defects. The verifier is not the worker prompt with "review" appended.

Phase 3: Bind to the host

Default to portable Markdown briefs, usable by an available agent or by flow-work directly. Check existing specialists before adding a redundant role. Generate native agent files only when the user requests them for an identified host; then read references/native-agents.md.

Phase 4: Exercise the specialist

When delegation is available and authorized, give the worker a small realistic task in an isolated fixture, and give a different verifier its raw result plus the contract. Keep expected answers out of the worker context. Fix only observed gaps in the briefs. Never touch production systems to test a prompt. Without an independent run, label the briefs untested.

Anti-patterns

  • A role that is "world-class senior engineer" plus generic best practices.
  • Installing or registering agents when only a reusable brief was requested.
  • Pinning a model or tool name without checking host support.
  • Relaxing repository gates so a generated specialist succeeds.

Done when

  • Both briefs are grounded in current repository evidence and task scope.
  • Write ownership, acceptance, stop conditions, and evidence outputs are explicit.
  • Validation and host-registration status are reported accurately.

Hand the briefs to flow-orchestrate for coordinated work or flow-work for a bounded direct task.

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.