Stats
Actions
Tags
Help us improve
Share bugs, ideas, or general feedback.
From claude-commands
Runs a 4-gate adversarial quality harness (/es, /er, /code_standards, independent review) that iterates on a goal until all gates pass. Useful as a final convergence step before declaring work done.
npx claudepluginhub jleechanorg/claude-commands --plugin claude-commandsHow this skill is triggered — by the user, by Claude, or both
Slash command
/claude-commands:goal_harnessThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Iterative goal-driven development with a 4-gate adversarial quality harness.
Reviews a goal against its definition-of-done and approves or rejects with a structured fix-list. Auto-fired by the goal skill or invoked on demand via /goal-judge.
Validates AI agent claims against evidence trail in coding workflows. Catches unsubstantiated 'done', 'tests pass', 'fixed' without proof like outputs, diffs, or logs. Auto-triggers on completion keywords.
Auto-loop execution workflow with quality gates. Use when starting any non-trivial implementation task. Provides automatic task decomposition, code implementation, testing (L1-L4), and iterative quality gates until completion. Invoke with /autoworker.
Share bugs, ideas, or general feedback.
Iterative goal-driven development with a 4-gate adversarial quality harness. The primary agent works on a goal, then validates the work through independent adversarial subagents that evaluate against specific standards.
Define a goal and iterate until 4 adversarial gates all pass:
/es — Evidence Standards/er — Evidence Review/code_standards — ZFC + ZFC-leveling + root-cause-first (3 lanes)/converge (the convergence loop)/es and /er must both pass before mergingSubagent (via the Agent tool) reads both evidence-standards skill layers and evaluates the diff and evidence artifacts.
Returns: PASS / WARN / FAIL with file:line evidence.
Subagent (via the Agent tool) runs evidence-review + evidence-standards synthesis against the diff and evidence artifacts (not the raw goal text).
Subagent reviews: Map claims → artifacts, rate STRONG/WEAK/MISSING.
Verdict: PASS / WARN / PARTIAL / FAIL / INCONCLUSIVE.
Normalization rule: WARN → PASS (document warnings). PARTIAL → FAIL (gaps remain). INCONCLUSIVE → FAIL (cannot confirm).
| Lane | Skill Path | Checks |
|---|---|---|
| ZFC | ~/.claude/skills/zero-framework-cognition/SKILL.md | No keyword routing, heuristic scoring, regex intent |
| ZFC-leveling | ~/.claude/skills/zfc-leveling-roadmap/SKILL.md | Level-up fields, modal scoping, stale flag guards |
| Root-cause-first | ~/.claude/skills/root-cause-first/SKILL.md | Root cause documented before adding protections |
Dispatch all 3 in parallel. Any lane FAIL → overall /code_standards FAIL.
Independent subagent reviews the full diff as a code reviewer. Looks for: bugs, anti-patterns, missing tests, security issues, dead code. Returns PASS or FAIL with file:line evidence.
| Raw verdict | Normalized | Harness treatment |
|---|---|---|
| PASS | PASS | Gate passes |
| WARN | PASS | Gate passes (document warnings) |
| PARTIAL | FAIL | Gate fails — evidence gaps remain |
| INCONCLUSIVE | FAIL | Gate fails — cannot confirm |
When running in cmux (background/scheduled mode):
scripts/goalexec-loop.sh for terminal-based loops (canonical)scripts/goalexec-loop-cmux.sh for cmux-aware loops (canonical)scripts/goal-loop.sh and scripts/goal-loop-cmux.sh are legacy aliases~/.goal-loop//goal — builtin Claude Code goal command/es — Evidence Standards (reference/display)/er — Evidence Review (adversarial synthesis)/code_standards — Coding standards dispatch (ZFC + ZFC-leveling + root-cause-first)/converge — Iterative goal achievement loop (formerly /goalexec)/converge_define — Define-only variant