agentleFS
soumatheusgomessoumatheusgomes/agentville/.claude/skills/goal-builder/SKILL.md

goal-builder

Skill9 starsChanged 3 months agoApache-2.0View on GitHub

Use when asked to generate, build, prepare, optimize or assemble a /goal command for autonomous or overnight execution — triggers: "assemble goal", "generate autonomous goal", "prepare overnight task", "goal for autonomous session", "optimize goal", "goal builder".

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is 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.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every condition needs a check whose result appears in the transcript (npm test exit 0, git status clean, printed count).

    line 15
  2. Distinctive:

    Always offer a recommended default in the options.

    line 65
  3. Distinctive:

    If a spec was generated: confirmed file path.

    line 89

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Goal Builder

Generates an optimized /goal for autonomous execution, with a 5-phase superpowers pipeline, transcript-verifiable checks, and a mandatory safety cap.

Invariants — never violate

  • The evaluator (Haiku) reads only the transcript. A check only proves something if the agent writes the output in the conversation.
  • Every condition needs a check whose result appears in the transcript (npm test exit 0, git status clean, printed count).
  • Max 4000 chars in the condition.
  • Superpowers pipeline always present — 5 phases. No user instruction = use default pipeline.
  • Goodhart Effect: the agent optimizes exactly what is measured. Rich task → reference an external spec with quality requirements.

Anatomy of a /goal (3 + 1)

#ComponentExample
1Measurable end state"all tests in test/auth pass"
2Transcript-verifiable check"npm test exits 0"
3Constraints"without altering existing test files"
+Mandatory cap"or stop after 80 turns"

Default superpowers pipeline (5 phases)

PHASE 1  superpowers:brainstorming               — design + spec (review if spec already exists)
PHASE 2  superpowers:writing-plans               — detailed implementation plan
PHASE 3  superpowers:executing-plans             — execute plan task-by-task with checkpoints
PHASE 4  superpowers:verification-before-completion — verify everything before finishing
PHASE 5  superpowers:finishing-a-development-branch — integrate: merge, PR or cleanup

Default cap: 80 turns. Calibrate: trivial task → 40t; broad task → 120t+. Minimum ~15t per phase.

When to replace: user lists alternative phases with explicit commands/skills → replace pipeline entirely.

MANDATORY Workflow

1. Collect requirements

  • Doc/spec provided → Read to extract objectives, criteria, constraints, paths. Do not re-ask what is already clear.
  • Record which pipeline to apply before proceeding.
  • Existing spec → PHASE 1 = "review/validate" (not recreate).
  • Existing plan → PHASE 2 = "validate/update" (not recreate).

2. Map gaps → AskUserQuestion (max 4 per call)

PointNotes
Objective / end stateHeart of the goal — without this there is no condition
How to prove (check)Output that appears in the transcript
Scope / pathsLimits where the agent operates
ConstraintsWhat must not break/change
Cap (turns)Default 80t. Reduce if trivial
Quality beyond the checkRequirements the check doesn't capture (anti-Goodhart)
PHASE 1 skip or review?Spec already exists → brainstorming in review mode
PHASE 2 update?Plan exists → validate existing vs. recreate?

Always offer a recommended default in the options.

3. Decide format

  • Simple task (1 end state, obvious check, few requirements) → inline /goal with default pipeline.
  • Complex task (multiple criteria, subjective quality, multi-domain, > 4000 chars) → spec + /goal that references it.
    • Existing spec pointed to → reference it, do not duplicate.
    • New spec → Write to docs/specs/<slug>.md.

4. Assemble the /goal

Mandatory checklist:

  • ✅ Measurable end state
  • ✅ Transcript-verifiable check
  • ✅ Explicit constraints
  • ✅ Cap calibrated to scope
  • ✅ 5-phase pipeline in exact order with "PHASE N OK" markers
  • ✅ Prohibition on skipping phases
  • ✅ do not ask questions, decide independently

5. Deliver

  • Copyable block with /goal ....
  • If a spec was generated: confirmed file path.
  • "Before running" checklist: auto mode on, git status clean, spec reviewed if it exists.

Templates

A) Simple task — inline goal

/goal <end state> proven by <check in transcript>; without <constraint>.

Execute the 5-phase superpowers pipeline in exact order, WITHOUT skipping any:
PHASE 1 superpowers:brainstorming — understand requirements, design and spec; print "PHASE 1 OK".
PHASE 2 superpowers:writing-plans — create detailed implementation plan; print "PHASE 2 OK".
PHASE 3 superpowers:executing-plans — execute plan task-by-task with checkpoints; print "PHASE 3 OK".
PHASE 4 superpowers:verification-before-completion — verify everything before finishing; print "PHASE 4 OK".
PHASE 5 superpowers:finishing-a-development-branch — integrate (merge/PR/cleanup); print "PHASE 5 OK".

End state proven: (1) <gate command> exiting 0; (2) all 5 "PHASE N OK" markers printed in order.
Do not ask questions, decide independently. Stop after <N> turns.

B) Complex task — spec + goal

docs/specs/<slug>.md:

# <Title> — Autonomous execution spec

## Objective
<desired result in 1-2 sentences>

## Acceptance criteria (each verifiable)
- [ ] <criterion 1> — proof: <command/output>
- [ ] <criterion 2> — proof: <command/output>

## Quality requirements (anti-Goodhart)
<what "good" means beyond the check: UX, business rules, edge cases>

## Scope
- Touch: <paths>
- DO NOT touch: <paths/constraints>

## Definition of done
<final state + all checks green + cap>

/goal line:

/goal implement everything in docs/specs/<slug>.md until all acceptance criteria pass, each proven by the indicated command.

Execute the 5-phase superpowers pipeline in exact order, WITHOUT skipping any:
PHASE 1 superpowers:brainstorming — review spec at docs/specs/<slug>.md (DO NOT recreate); print "PHASE 1 OK".
PHASE 2 superpowers:writing-plans — create/validate plan based on spec; print "PHASE 2 OK".
PHASE 3 superpowers:executing-plans — execute plan task-by-task; print "PHASE 3 OK".
PHASE 4 superpowers:verification-before-completion — run <gate command>; verify coverage; print "PHASE 4 OK".
PHASE 5 superpowers:finishing-a-development-branch — integrate; print "PHASE 5 OK".

End state proven: (1) <gate command> exiting 0; (2) all 5 "PHASE N OK" markers printed in order; (3) git status clean.

Quality (anti-Goodhart): <requirements the gate does not capture>.
Constraints: <what must not change>.
Do not ask questions — decide independently. Stop after <N> turns.

Anti-patterns

BadGood
Goal without pipeline5-phase superpowers pipeline always present
No capor stop after N turns always
Check the evaluator cannot seeCheck output printed in transcript
PHASE N OK markers absentOne marker per phase — evaluator proves execution
Re-asking what is in the docRead doc, extract, ask only what is missing
Undersized cap5 phases × ~15t minimum = 80t default
Pipeline omitted for "simple task"Pipeline is DEFAULT — always present, always
New spec in tasks/New spec goes in docs/specs/<slug>.md

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.