Verify Against Rubric

---
name: verify-against-rubric
description: |-
  Spawn a stateless sub-agent (Claude Code `Agent` tool, Cursor agent mode, or equivalent) with ONLY the artifact path plus the locked rubric plus read-only tools. Sub-agent returns a structured verdict (satisfied, needs_revision, or failed) plus per-criterion feedback. Distinct from self-critique-and-revise (same-context; rewrites in place). Use after pr-package, contract-signoff, or implement-approved-slice on HIGH-complexity slices when independent verification matters more than throughput. Not as a routine pass on LOW or MEDIUM slices (self-critique-and-revise covers those at lower cost), but always when approve-plan (ADR-0208) or a zero-finding review-hard verdict (ADR-0145) dispatches it. Do not use when no locked rubric exists yet or the rubric is vague (run resolve-contract-gaps first), or when no active task folder exists yet (run task-init first). For 4 or more artifacts against one rubric, use verify-against-rubric-fleet.
metadata:
  category: "planning-and-validation"
  primary-cursor-mode: "Plan"
  multi-repo-aware: "false"
  context-layers-consumed: "memory"
  context-layers-produced: "memory"
  tools: "Read, Write, Edit, Bash, Glob, Grep"
  x-wos-profiles: "minimal, core, full"
  provenance: "first-party"
  suggested-model: "claude-sonnet-5-5"
---

Act as a senior/staff engineer dispatching an independent stateless verification sub-agent against a locked rubric.

Goal:
Verify a task artifact (PR package, signed-off contract, executed slice) against a locked rubric using an isolated sub-agent context. The sub-agent sees only the artifact + the rubric (no TASK_STATE.md, no DECISIONS.md, no prior conversation). Sub-agent returns a structured verdict; main thread persists it to `VERIFICATION_LOG.md` and emits the next-step handoff. Closes the same-context bias gap that `self-critique-and-revise` (in-thread) inherits. Per ADR-0033 and Anthropic Outcomes (2026-05-06; +10pp success vs same-context critique).

Mandatory context bootstrap (before any output):
- Read these sections in `WORKFLOW_OPERATING_SYSTEM.md` first:
  - `## LLM execution contract`
  - `## Editor mode policy` (mode definitions only; the tool mapping table is lazy-loaded in `wos/editor-mode-mappings.md` and needed only for non-Claude-Code tools)
  - `## Global output contract` (including **Adaptive handoff** and **Mode selection rule**)
  - `## Cross-cutting workflow guardrails`
- **Bootstrap tiers:** the light-weight commands (`branch-commit`, `what-next`, `where-we-at`, `slice-closure`, `compact-task-memory`) plus the high-frequency `implement-approved-slice` and `sync-task-state` (v3 wave1 item D) read the four sections above with two subsections of `## Cross-cutting workflow guardrails` skipped: `### External web access (centralized)` and `### Sequencing heuristics (by phase)`. Everything else is read at every tier, including `### Proposal vs approved persistence` and `### Substrate peer ownership (per ADR-0034)`, since all seven write substrate sections and reason about PROPOSED (`state-reconcile` stays on the full tier for cross-artifact judgment). The full tier is measured at 12109 tokens, the four always-read sections combined; the two skipped subsections are 1,034 of those (measured 2026-09-28), so the reduced tier is about 11,074. The leaf-reviewer tier (`verify-against-rubric`, ADR-0226) reads only `## Global output contract`, measured at 5215 tokens, plus its rubric.
- **Session bootstrap reuse (skip-if-unchanged; v3 wave1 item D):** WHEN this conversation already read the bootstrap sections in an earlier turn still VISIBLE in the context window AND `WORKFLOW_OPERATING_SYSTEM.md` has not changed since, the command MAY skip the re-read and cite the earlier one, emitting one Command transcript line: `Bootstrap: reusing turn <N> read, WOS unchanged`. Scoped exception to the context-budget re-fetch rule (`wos/context-budget.md`, "The re-fetch rule"), because these sections are one large, static, byte-identical read repeated every turn; every other tool result still re-fetches. VISIBLE means the section text itself is still present and quotable now, not merely that a record of the earlier read exists: a harness that clears a tool result while the record survives (ADR-0114) has not satisfied VISIBLE, and self-declared memory after a compaction never qualifies. A stateless-per-turn harness is excluded. The transcript line is mandatory; a silent skip is invalid output.
- **Resolving the spec and `wos/<topic>.md`:** canonical repository FIRST. Otherwise use active-adapter docs: `~/.claude/workflow-docs` (`CLAUDE_WORKFLOW_DOCS_DIR`), `~/.cursor/workflow-docs` (`WORKFLOW_DOCS_DIR`; project `.cursor/workflow-docs` on project installs), `~/.codex/workflow-docs` (`CODEX_WORKFLOW_DOCS_DIR`), or `<kimi-home>/workflow-docs` (`KIMI_WORKFLOW_DOCS_DIR`). Nonempty overrides win. Kimi home: `KIMI_CODE_HOME`, else existing `~/.kimi-code`, else existing `~/.kimi`, default `~/.kimi-code`. Use one root for both paths; try other adapters if absent or unknown. Report resolved root, fallback to other adapters, or missing mandatory context in `### Command transcript`. Installed snapshots refresh and prune retired topics only on sync.
- Read additional sections only when relevant to this command's role.
- Align all routing recommendations and next-command suggestions with the current command set.
- **Official next-command names only:** recommendations, including `Run now`, use registered names from `commands/<name>.md` or `commands/<name>/SKILL.md`; installed mode uses the available Fhorja command or skill inventory. No invented aliases. `Run now: none` with `Mode: N/A` ends only if no honest next step exists, defined under `### Official command names (routing integrity)` (ADR-0126).
Required inputs:
- artifact path (PR_PACKAGE.md, contract document, slice file, etc.)
- rubric source: EITHER inline rubric (criteria list with verdict thresholds) OR reference to a section in IMPLEMENTATION_PLAN.md or DECISIONS.md (path + section anchor)
- task folder path
- optional: prior verdict id (for delta check vs previous run)

Task repository files to update:
- VERIFICATION_LOG.md (new optional task file; append-only)
- TASK_STATE.md (`## Last completed step` references the new verdict id)

Operating rules:
- **Handoff:** end with the adaptive `### Handoff` block per `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract` (Mode A compact, Mode B full, or Mode C parallel-fanout when triggered).
- **Bootstrap tier: leaf reviewer (ADR-0226).** This command runs on the leaf-reviewer tier of the bootstrap block above: from `WORKFLOW_OPERATING_SYSTEM.md` it reads only `## Global output contract`, plus its rubric, whether a user runs it or `approve-plan` (ADR-0208) or `review-hard` (ADR-0145) dispatches it. It skips `## LLM execution contract`, `## Editor mode policy` and `## Cross-cutting workflow guardrails`, which govern an agent that edits the task or dispatches other commands. This one edits no code, plan or decision. It writes only its own verdict entry and the one TASK_STATE.md line, by the rules in Steps 4 and 5, and its one dispatch is the stateless reader Step 2 defines. A rule from a skipped section that this command needs is written here or in the shared blocks below, not left to the skipped read. The tier changes what this command reads from the spec. It never changes what the Step 2 dispatch carries: the artifact and the rubric, nothing else (ADR-0033, ADR-0145).
- **Step 1: Lock the rubric.** Either accept the inline rubric as-is or extract the referenced section from IMPLEMENTATION_PLAN.md / DECISIONS.md. The rubric MUST list discrete criteria (each verifiable independently). Refuse with NO_OP_TRACE if the rubric is vague ("looks good", "feature works") -- route to `resolve-contract-gaps` first.
- **Step 2: Spawn sub-agent (the load-bearing step).** Invoke the host's stateless sub-agent primitive (Claude Code `Agent` tool with `subagent_type: general-purpose`; Cursor agent mode subagent; Codex agents) as a `judgment` dispatch (ADR-0236): apply the judgment row of `wos/model-routing.md ## Dispatch roles`, passing its model per call, and write `effort: inherited from session` in `### Command transcript` when the path takes no effort. The verdict is the deliverable, so it is never routed down, whether a person, `approve-plan` or `review-hard` dispatched this run. The sub-agent receives ONLY:
  - The artifact file path (read-only)
  - The rubric (verbatim)
  - Instruction: "Verify the artifact against each criterion. Return a structured verdict: per criterion, mark satisfied / needs_revision / failed with a 1-2 line reason. Then an overall verdict."
  - NO TASK_STATE.md, NO DECISIONS.md, NO prior conversation history.
  - A rubric MAY quote cited D-N authorization and P-N provisional entries verbatim, as `approve-plan` does. The quoted text is part of the rubric; the full decisions file is never passed. Evidence pointers permit narrowly checking canonical branch code or verbatim brief quotes, never opening investigation carriers or author recommendations; report unknown support as unverified.
- **Step 3: Receive verdict.** Sub-agent returns the verdict in the agreed structured format. Main thread does NOT re-evaluate or override the verdict; it persists it.
- **Step 4: Persist to VERIFICATION_LOG.md.** Append a new entry with: verdict id (auto-incremented), date, artifact path, rubric source, per-criterion verdict, overall verdict, sub-agent identifier (Claude Code Task ID or equivalent), and the timestamp.
- **Step 5: Update TASK_STATE.md.** Per the canonical 5-section write pattern in `commands/_shared/task-state-slice-closure-pattern.md`, update `## Last completed step` to reference the new verdict id.
- **Step 6: Emit handoff.** If verdict = satisfied: route to the next planned step (often `pr-package` submission or the next slice). If needs_revision: route contract wording revisions to `contract-signoff`, unresolved contract gaps to `resolve-contract-gaps`, and packaging revisions to `pr-package`. Use `implement-slice-complement` only for bounded implementation fixes with an existing anchor slice and `IMPLEMENTATION_PLAN.md`; if that anchor or plan is missing, route to `implementation-plan`. For other artifacts, route to their owning command. Include verdict feedback inline. If failed: route to `direction-adjust` or `decision-interview` based on the failure type.
- Do NOT run this command as a routine pass on LOW or MEDIUM complexity slices: `self-critique-and-revise` is the cheaper in-thread alternative. The exception is a dispatch: `approve-plan` runs it on every plan it approves (ADR-0208; the one-slice route skips approval and runs `check-doc-sync.sh --against HEAD` instead, ADR-0225) and a zero-finding `review-hard` verdict runs it on product code (ADR-0145), whatever the complexity.
- Do NOT spawn sub-agent without an explicit locked rubric. Vague rubrics produce vague verdicts.

Required output:
1. Rubric source (inline or referenced section)
2. Number of criteria evaluated
3. Per-criterion verdict (table)
4. Overall verdict (satisfied / needs_revision / failed)
5. Sub-agent identifier (Task ID or equivalent)
6. Verdict id in VERIFICATION_LOG.md
7. Recommended next command based on verdict

### Review prompt scaffold (optional)
When the review directives in this command are ambiguous, parse them in three labeled parts: Instructions (what to do), Context (background, not a rule), and Constraints (hard limits that override the rest). This separation is optional and adds signal only where reviewers report ambiguity; do not tag mechanically or let it bloat the prompt.
### Claim grounding (active epistemic humility)
**Claim grounding (active epistemic humility).** This block governs what you may assert and how you record it. It is keyed to the substrate section you are writing, not to which command is running, and it is INERT on any output that writes none of the claim-bearing sections below. Full contract and rationale: `wos/active-epistemic-humility.md`.

1. When this applies. This block fires ONLY while you are writing a claim-bearing substrate section: `TASK_STATE.md ## Current known facts`, `## Risks to watch`, `## Observations`, `## Active files in scope`, `## Canonical decisions`; `DECISIONS.md ## Locked decisions`; `IMPLEMENTATION_PLAN.md ## Current gaps`, `## Risks and mitigations`; `IMPACT_ANALYSIS.md`; `EXTERNAL_RESEARCH.md`; `REFERENCES.md`; or any section whose content is a statement a later command or a human decision will act on. WHEN your output writes none of these, this block imposes nothing: skip it and proceed. This is the D-13 inert clause; a fully-grounded or claim-free output pays nothing.

2. The unit is the load-bearing claim. A load-bearing claim is one a downstream command or a human decision consumes. A passing aside is not load-bearing; a statement someone will act on is. Apply the rest of this block per load-bearing claim, not per sentence.

3. Ground it or abstain. Before you assert a load-bearing claim, trace it to the enumerable grounded set: a captured `REFERENCES.md` entry, a file read in this session, command output actually seen, or a passing deterministic gate. A claim supported only by model memory is OUTSIDE the grounded set, including when you are right, because that support is not observable. WHEN a load-bearing claim falls outside the set, do NOT assert it: either investigate until it is grounded, or abstain per rule 6.

4. Status records provenance, never confidence. WHERE you attach an epistemic status to a claim, the status names WHERE THE CLAIM CAME FROM: a `REFERENCES.md` entry title, a file path plus line, or the gate output it came from. It SHALL NOT express a degree of certainty. Do NOT add a confidence field, a numeric threshold, or a self-assessment prompt anywhere; a self-reported confidence signal is not a usable control signal (`wos/active-epistemic-humility.md` Part 1.3). A status whose referent slot is empty is read as UNKNOWN, not as a weak yes.

5. Persisted claims carry the status; chat-only claims carry it when they route. Every load-bearing claim you write into a task-memory artifact carries its provenance referent, and that referent travels with the claim so a later command reads it too; do not drop it at the write boundary. A load-bearing claim that appears only in a chat-turn output carries a status only when it crosses the grounding boundary and triggers a route (an abstention, an escalation).

6. Abstain as a routed continuation, never a bare refusal. WHEN you abstain, name the specific investigation that would settle the question AND route to the command that runs it (`capture-references`, `code-locate`, `incident-triage`, or the fitting one). A withholding that stalls the work is invalid output. Abstention is distinct from `NO_OP`: `NO_OP` means there is no work to do; abstention means there is work and the grounding to do it is missing.

7. An unfired gate is not evidence. The absence of a fired check does not mean grounding existed. Do not read silence here as a pass.
### Standard output layout (required)
Produce the command output using this structure (English only):

### Artifact changes
Follow `## Global output contract` in `WORKFLOW_OPERATING_SYSTEM.md` for `APPLIED` / `PROPOSED` / `SKIP` rules. Every listed file carries one of those three tokens, in Lean output too; a prose verb like "written" is not a label.

### Command transcript
Brief audit trail (max 4 lines; max 3 in no-op runs with `NO_OP_TRACE`).

### Handoff
Use the adaptive ending format of `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract`. Every Handoff is one fenced `text` block with all four lines, `Run now:`, `Mode:`, `Work complexity:` and `Reason:`, on a stop and on a refusal too; the terminal form is `Run now: none` with `Mode: N/A`. A new `Mode:`, a model or a fresh session is never a stop, an offered choice is one, and `Reason:` names a role, never a model.
### Definition of done (command output)
- Rubric is locked and discrete (criteria are independently verifiable).
- Sub-agent invoked with isolated context (no TASK_STATE.md, no DECISIONS.md, no prior history).
- Verdict returned with per-criterion breakdown and overall classification.
- VERIFICATION_LOG.md updated with the verdict entry.
- TASK_STATE.md `## Last completed step` references the verdict id.
- Handoff routes based on verdict: satisfied -> next step, needs_revision -> the artifact's owning command per Step 6, failed -> direction-adjust or decision-interview. A needs_revision route to `implement-slice-complement` requires an existing anchor slice and `IMPLEMENTATION_PLAN.md`.
- Before declaring this output done, confirm it satisfies the shared **Definition of done (command outputs)** and **Gate conditions** in WORKFLOW_OPERATING_SYSTEM.md.

Quality bar:
The verdict is only as good as the rubric. If the rubric is vague, the verdict is vague. Refuse vague rubrics. The sub-agent isolation is the load-bearing differentiator vs `self-critique-and-revise`; do not collapse them.