workflow-investigate-review

name	workflow-investigate-review
description	Review Smelter investigation artifacts for evidence gaps, contradictions, and unsupported claims.
version	0.55
type	workflow
consumes	investigation.md
produces	investigation-review.md
default_pattern	A
default_agent	explore-high
supports_patterns	["A"]
result_types	["pass","fail","reshape"]
min_verification_rounds	2
verification_rounds	[{"n":1,"focus":"omission","prompt_template":"templates/verification/round-1-omission.md"},{"n":2,"focus":"contradiction","prompt_template":"templates/verification/round-2-contradiction.md"}]
gate	{"postcondition":[{"file_exists":"investigation-review.md"},{"contains_decision":"pass|fail|reshape"}]}

workflow-investigate-review

Overview

Reviews investigation.md. Identifies missing coverage, excess, and risks before the active mode's next planning or execution stage consumes it.

Core principle: Downstream stages only get evidence that survives 2-round review. Partial review is not review.

Violating the letter of this rule is violating the spirit of this rule.

Announce at start: "I'm using workflow-investigate-review to run 2-round verification on investigation.md."

The Iron Law

NO DOWNSTREAM HANDOFF WITHOUT 2/2 REVIEW ROUNDS AT pass

Skill-level pass requires completed_rounds === 2 && all rounds result === pass. Declaring pass with completed_rounds < 2 is blocked by hook.

Output

investigation-review.md:

• ## Verdict — pass / fail / reshape
• ## Coverage Check — whether any area is missing
• ## Risks Validated — re-validated risks
• ## Reshape Target — upstream skill on reshape (e.g., workflow-brainstorm)

Red Flags - STOP

Thought	Reality
"The investigation looks fine, skip round 2"	Hook blocks sub-2 pass. Run both rounds.
"No contradictions, round 2 is wasted"	Round 2 finds the contradictions. Absence is what the round discovers.
"Reshape feels heavy, downgrade to fail"	If scope is wrong, reshape. Calling reshape fail wastes the investigate re-run without fixing the upstream brainstorm.

Rationalization Prevention

Excuse	Reality
"Coverage is good enough"	Reviewer decides, not implementer. Record the verdict.
"Risks are already in brainstorm"	## Risks Validated is a fresh re-evaluation, not a copy.
"Just this once"	No exceptions. Iron Law applies to every run.

Routing

• pass → next skill in the active mode's default order
• fail → re-run workflow-investigate (a specific area may be designated)
• reshape → workflow-brainstorm (scope was wrong from the planning stage; evidence required)

Mode-specific exits

• /brainstorm mode: on pass, workflow-tasker
• /implement mode: on pass, workflow-implementation-plan
• /fix mode: on pass, workflow-write-test
• /explore mode: on pass, mode_transition gate

Multi-Pass Verification (2-Round Enforcement)

This skill runs 2 mandatory rounds before declaring pass. Each round has a distinct focus:

Round	Focus	Question
1	Omission	Is any required area (schema, API surface, UI, security, docs) missing?
2	Contradiction	Are there conflicting findings across areas or with the brainstorm?

Agent assignment per round

• Pattern A: Prefer different agent types across rounds. If reusing the same type, reset prompt context per round (fresh perspective).
• Pattern B: Each round runs with 95% consensus (3 × N agents × consensus rounds).
• Pattern D: Lead orchestrates 2 rounds, assigning different viewpoint sub-agents.

State recording

All rounds recorded in state.json.team_runtime.workflow-investigate-review.rounds[]. Skill-level pass is declared only when completed_rounds === 2 && all rounds result === pass.

Failure handling

• Any round fail → skill-level fail with cause: verification_failed, evidence: {round, focus, findings[]}
• Producer-chain routes to workflow-investigate (the consumed artifact's producer)
• On re-entry after upstream fix, both rounds re-run (not just failed ones)

Anti-evasion enforcement

1. Do not inject prior round conclusions into current round prompts (bias prevention)
2. Critic Watchdog blocks "already verified" skip statements
3. Declaring pass with completed_rounds < 2 is blocked by hook

Terminal State — Required Next Skill

REQUIRED NEXT SKILL on pass:

• /brainstorm mode → workflow-tasker
• /implement mode → workflow-implementation-plan
• /fix mode → workflow-write-test
• /explore mode → mode_transition gate (may route to /brainstorm, /implement, or /fix per suggested next mode)

On fail: route back to workflow-investigate. On reshape: route back to workflow-brainstorm.

Do NOT:

• Invoke workflow-coding, workflow-write-test, or any implementation skill
• Stop on pass, report completion, or ask "shall I continue?"
• Offer A/B/continue choices — Iron Law #1 forbids pausing at non-human-check stages

Evidence Integrity (Mechanical Enforcement)

Every fail verdict MUST cite at least one anchor in the strict form:

Evidence: path/to/file.ext:LINE[-LINE] "verbatim quote substring"

Rules:

• The path must exist on disk at this session's cwd.
• The quoted substring must appear on the cited line (or within the range).
• Quote is a substring match after whitespace normalization — not a regex.
• A PostToolUse hook (scripts/review-evidence-verifier.mjs) blocks writes that violate these rules.
• Do not paraphrase — quote the line verbatim.
• If you cannot produce a verified anchor, the symptom is not grounded; do not emit a fail verdict.