跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

completion-gate

Validates agent claims against evidence trail. Use when verifying an agent has actually done what it claims — auto-fires at workflow end. Catches 'done' without proof, 'tests pass' without output, 'fixed' without verification. Called by cook and team.

Rune-kit · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add Rune-kit/rune --skill completion-gate

技能说明


name: completion-gate description: "Validates agent claims against evidence trail. Use when verifying an agent has actually done what it claims — auto-fires at workflow end. Catches 'done' without proof, 'tests pass' without output, 'fixed' without verification. Called by cook and team." user-invocable: false metadata: author: runedev version: "1.10.0" layer: L3 model: sonnet group: validation tools: "Read, Bash, Glob, Grep"

completion-gate

Purpose

The lie detector for agent claims. Validates that what an agent says it did actually happened — with evidence. Catches the #1 failure mode in AI coding: claiming completion without proof.

Triggers

  • Called by cook in Phase 5d (quality gate)
  • Called by team before merging stream results
  • Called by any skill that reports "done" to an orchestrator
  • Auto-trigger: when agent says "done", "complete", "fixed", "passing"

Calls (outbound)

None — pure validator. Reads evidence, produces verdict.

Called By (inbound)

  • cook (L1): Phase 5d — validate completion claims before commit
  • team (L1): validate cook reports from parallel streams

Execution

Step 1 — Collect Claims

Parse the agent's output for completion claims. Common claim patterns:

CLAIM PATTERNS:
  "tests pass" / "all tests passing" / "test suite green"
  "build succeeds" / "build complete" / "compiles clean"
  "no lint errors" / "lint clean"
  "fixed" / "resolved" / "bug is gone"
  "implemented" / "feature complete" / "done"
  "no security issues" / "sentinel passed"

Extract each claim as: { claim: string, source_skill: string }

Step 1a — Type Each Claim (Claim Discipline)

Before hunting for evidence, type the claim by the grammar it was written in. Hallucination is an unverified claim wearing the grammar of an observation — the grammar is the tell, and it is readable in the sentence itself.

TypeMeaningGrammar it may wear
OBSERVEDSeen this session: ran it, read it, measured it"X is / does / returns …"
DERIVEDFollows from OBSERVED facts via a statable mechanism"X should / will / implies …" + the why
PRIORTraining knowledge, may be stale"X is typically … / was, as of …"
ASSUMEDUnverified and required by the conclusion"I am assuming X — if wrong, then …"

This changes what the gate is looking for in Step 2:

  • OBSERVED → demands an evidence artifact. No artifact = FAIL. This is the existing gate.
  • DERIVED → demands the mechanism be stated, and its OBSERVED inputs to be present.
  • PRIOR / ASSUMED → not a failure. A claim honestly delivered as assumed is the correct output when the check was not run. Record it as an open item; never score it as a lie.

Claims are promoted only by tools — checking a PRIOR makes it OBSERVED. Restating it more confidently does not. Confidence that grew from effort, repetition or fluent prose resets to the last evidence-backed level.

Step 1b — Stub Detection (Existence Theater Check)

Before checking claims, scan all files created/modified in this workflow for stubs:

Grep for stub patterns in new/modified files:
- "Placeholder" | "TODO" | "Not implemented" | "NotImplementedError"
- Functions with body: only `return null` / `return {}` / `pass` / `throw`
- Components returning only a single div with no logic

If ANY stub detected:

  • Add synthetic claim: "implemented [filename]" → CONTRADICTED (file is a stub)
  • This catches agents that create files but don't implement them

Step 1c — Self-Validation Check

If the skill that just ran has a ## Self-Validation section, extract its checklist and treat each item as an implicit claim:

For each Self-Validation check in the skill's SKILL.md:
  1. Read the check (e.g., "at least one assertion per test")
  2. Look for evidence in tool output that this check was satisfied
  3. If evidence found → add as CONFIRMED claim
  4. If no evidence → add as UNCONFIRMED claim ("Self-Validation: [check] — no evidence")

Why: Self-Validation catches domain-specific quality issues that generic claim matching (Step 2) cannot detect. A test skill knows "no assertions = useless test" but completion-gate doesn't — unless the skill's Self-Validation tells it to check.

Step 1d — Execution Loop Audit

Before validating claims, audit the agent's tool call pattern for execution loops that indicate the agent was stuck but didn't report it:

Classify the agent's tool calls from this workflow into two categories:

CategoryToolsExpected in Phase 4
ObservationRead, Grep, Glob, Bash(grep/ls/cat)<40% of calls
EffectWrite, Edit, Bash(build/test/npm)>60% of calls

Loop patterns to detect:

PatternDetectionVerdict Impact
Observation chain: 6+ consecutive observation tools in Phase 4Count longest observation-only streakAdd WARN: "Agent had {N}-call observation streak during implementation — possible analysis paralysis"
Low effect ratio: <20
阅读完整来源 ↗