evaluator

Plugin evaluator from marketplace forge-studio in ArtMin96/forge-studio. Commands read from the repo's manifest — 1 month ago.

Skills (14)

• assess-proposal

  Adversarial review of a self-evolution proposal artifact. Emits a pass/fail verdict against four criteria. Pure read — never mutates harness files.

• challenge

  Use whenever the user wants a deep critique of a recently completed feature — runs Stage 1 self-review against the diff, then Stage 2 git-history verification (does the change actually pass historical evidence, not just current tests?). Forks to a general-purpose agent so the critique is independent

• devils-advocate

  Use when the user asks to "play devil's advocate", "argue the other side", "what could go wrong here", or wants the strongest possible counter-case against a design decision before committing — produces a structured tradeoff analysis with the opposing view's evidence.

• gate-report

  Use when the user wants a single consolidated view of every quality warning the hooks raised this session — duplicate reads, oversized outputs, scope-guard denials, secret-scan flags, drift checkpoints — grouped by severity with the offending file/line.

• grill-me

  Use when the user says "grill me", "interview me", "stress-test this plan", or you're in plan mode and the design has branches with unresolved decisions — runs a structured interview, walking each branch of the decision tree, recommending an answer for every question, and refusing to move on until s

• healthcheck

  Use when the user wants a one-command quality snapshot — auto-detects PHP and/or JS/TS projects and runs the appropriate pipeline (composer install + pest + pint + phpstan; npm install + tests + lint + typecheck). Returns a single PASS/WARN/FAIL summary plus the offending tool's output.

• optimize-description

  Run a 5-iteration description-optimization loop for a skill. Splits a query corpus 60/40 train/val, measures trigger rates, refines the description toward higher validation pass rate, and emits a proposal with the best-iteration result.

• postmortem

  Use after fixing a non-trivial bug, or after a `/safe-mode` exit, or whenever the user wants to convert a recent failure into a durable lesson — runs a structured autopsy with root-cause, bug classification (logic/race/state/env/integration), and a "could this have been caught earlier?" answer that

• prediction-audit

  Join SEPL proposal predictions against post-commit trace observations and report per-resource prediction error. Pure read — never mutates harness files or the ledger.

• run-evals-bench

  Run comparative benchmarks for a skill — with-skill vs without-skill, N iterations each. Emits per-iteration benchmark.json with pass_rate/time/token mean+stddev+delta, and per-run grading.json with per-assertion evidence.

• run-evals

  Validate eval JSON files for structural conformance against the per-skill evals/ convention. Parses each file, checks required fields and types, and emits a human-readable checklist of declared expectations.

• score-rubric

  Aggregate weighted criterion scores into a single rubric result. Reads a rubric JSON and a flat criterion-scores JSON, validates weight sum, and emits a result conforming to result.schema.json.

• verify-refs

  Cross-check file paths, function names, and external references claimed in your prior turn against the actual repo. Catches fabricated references (hallucinated file paths, misspelled symbols, dead URLs) before they reach a commit message, PR body, or user-facing summary.

• verify

  Use whenever a task is about to be marked done — runs the listed verification commands (tests, lint, type-check, behavioral spot-check), captures the actual output, and compares against `.claude/features.json` or the `/contract` criteria. Refuses to mark done unless every gate produced evidence; clo

Subagents (1)

• adversarial-reviewer

  Reviews code with a skeptical eye. Asks hard questions about edge cases, failure modes, and hidden assumptions. Use for security-sensitive or complex code.

Hooks (10)

• PostToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/php-static-analysis.sh

• PostToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/js-static-analysis.sh

• PostToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/test-nudge.sh

• PostToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/test-nudge-reset.sh

• PostToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/filter-test-output.sh

• PreToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/pre-commit-gate.sh

• PreToolUse

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/docs-drift-gate.sh

• TaskCompleted

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/task-completion-gate.sh

• SubagentStop

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/auto-verify.sh

• PostToolUseFailure

  bash ${CLAUDE_PLUGIN_ROOT}/hooks/route-failure.sh

Similar plugins