methodology-check

name	methodology-check
description	For any artifact ID, report which of the EPIC-001 4-layer pipeline layers (S10 FPF design, S11 BMAD quality gate, S12 OpenSpec structure, S13 Forgeplan automation) are satisfied and which need work. Also checks C4 coverage for ≥3-module architectural decisions. Returns a per-layer score 0-2 + aggregate percentage + concrete action items per gap. Read-only — never auto-fixes, never writes to forgeplan. Use before activating any Standard+ artifact as a final sanity check. Triggers: "methodology check", "check methodology", "4-layer coverage", "layer report", "/methodology-check", "проверь методологии", "layer coverage", "pipeline check", "coverage report", "method check"
origin	forgeplan

/methodology-check — 4-Layer Pipeline Coverage Report

For any forgeplan artifact, this skill reports which of the EPIC-001 methodology layers are satisfied and which need attention. The output is a structured markdown table with a per-layer 0-2 score plus concrete action items for every gap found.

Read-only contract: this skill never calls forgeplan_update, never writes files, and never dispatches other agents. It reports; the user or orchestrator decides what to do.

Foundation: EPIC-001 «4-Layer Pipeline Enforcement (S10→S13)» — Sprints Z6-Z10.

---

When to use

• Before calling forgeplan_activate on any Standard+ artifact — final sanity check.
• When a PR reviewer or guardian flags methodology gaps — get a diagnostic.
• Periodic audit: run against a set of active PRDs to find which need backfill work.
• User asks «is PRD-NNN fully covered?», «does this ADR meet all layers?», «methodology check».

When NOT to use

• Tactical-depth artifacts — only S12 + S13 apply; N/A is already the correct answer for S10/S11.
• Auto-fixing: if a gap is found, dispatch the appropriate agent (see action items in the report).
• Checking test coverage, code coverage, or linting — that is tester / code-reviewer territory.

---

Procedure

Step 1 — Parse target

Accept the artifact ID from the skill invocation argument or from the user's message. Supported kinds: PRD-NNN, RFC-NNN, ADR-NNN, EPIC-NNN, SPEC-NNN.

If no ID is provided, ask the user:

«Which artifact would you like to check? (e.g. PRD-061, ADR-007, RFC-002)»

Step 2 — Fetch artifact

artifact = forgeplan_get(id=<target>)

Extract and record:

• artifact.status — draft / active / stale / superseded / deprecated
• artifact.depth — tactical / standard / deep / critical
• artifact.body — full body for inline parsing
• artifact.links — all linked artifact IDs (via the links field or by parsing ## Related Artifacts in body)
• artifact.kind — prd / rfc / adr / epic / spec

If forgeplan_get fails (artifact not found) → stop and report:

ERROR: Artifact <target> not found in the forgeplan workspace.

Step 3 — Identify applicable layers

Build the layer checklist based on depth and kind:

Layer	When applicable
S12 OpenSpec (structure)	Always — every artifact
S13 Forgeplan (automation)	Always — every artifact
S10 FPF (design)	depth = standard / deep / critical
S11 BMAD (quality gate)	depth = standard / deep / critical
C4 (architecture extension)	ADR or RFC kind, AND body contains ≥3 distinct module references
RIPER (production-bug discipline)	Bug-fix artifact (EVID/PRD whose body, title, or parent links indicate production bug / incident / regression / race condition / SEV-1/SEV-2/P0/P1) AND depth is non-tactical

For tactical depth: mark S10, S11, and RIPER as N/A in the report (not required).

Record the applicable layer set. This drives Steps 5-10.

Step 4 — Fetch linked EVIDs

From the artifact links (or by scanning the body for EVID-NNN patterns), collect all referenced evidence IDs. Fetch each:

for evid_id in linked_evid_ids:
    evid = forgeplan_get(id=evid_id)
    # record body for S10 + S11 parsing

If no EVIDs are linked at all → S10=0 and S11=0 immediately (skip detailed checks for both).

Step 5 — S10 FPF check (design layer)

Skip if: depth=tactical (mark N/A).

Look in each linked EVID body for ADI hypothesis markers:

• ## Hypotheses considered
• ### H1 / ### H2 / ### H3 (or ### Hypothesis 1 etc.)
• **H1:** / **H2:** / **H3:** bold-pattern inline
• forgeplan_reason output block (heading ## ADI Cycle or ## FPF Reasoning)

Count distinct hypothesis mentions across ALL linked EVIDs. A hypothesis is distinct if it describes a different option/approach, not just a repeated mention.

Scoring:

Hypotheses found	Score	Status
≥3 distinct hypotheses in any EVID	2	✅ SATISFIED
1-2 hypotheses found	1	⚠️ PARTIAL
0 hypotheses found	0	❌ BLOCKER

Record which EVID (if any) contains the hypotheses.

Step 6 — S11 BMAD check (quality gate layer)

Skip if: depth=tactical (mark N/A).

Among all linked EVIDs, look for Profile B reviewer evidence. Identify a Profile B EVID by any of these markers in the EVID body:

• Text «Profile B reviewer» or «Profile B»
• Title containing «BMAD review», «adversarial review», «quality review»
• Bold verdict pattern **Verdict**: PASS / **Verdict**: CONCERNS / **Verdict**: BLOCKER

For each candidate Profile B EVID, check:

1. Does the body contain a ## Findings section?
2. Does it have ≥1 line item under ## Findings?
3. Is a verdict (PASS / CONCERNS / BLOCKER) explicitly stated?

Scoring:

Condition	Score	Status
≥1 Profile B EVID + ## Findings (≥1 item) + explicit verdict	2	✅ SATISFIED
Profile B EVID exists but ## Findings is empty or absent	1	⚠️ PARTIAL
No Profile B EVID found at all	0	❌ BLOCKER

Record the EVID ID and verdict if found.

Step 7 — S12 OpenSpec check (structure layer)

Count inbound + outbound links for the target artifact. Links can come from:

• The links field in the forgeplan_get response
• The ## Related Artifacts table in the body
• Any EPIC-NNN, PRD-NNN, RFC-NNN, ADR-NNN reference with a relation word («refines», «informs», «based_on», «supersedes», «parent»)

Delta-spec check (for superseded artifacts or artifacts with a supersedes link): Look for ## Delta-spec in the body — or the four sub-sections:

• ### ADDED
• ### MODIFIED
• ### REMOVED
• ### UNCHANGED

Scoring:

Condition	Score	Status
≥1 link present AND (no supersede, OR supersede + ## Delta-spec present)	2	✅ SATISFIED
≥1 link present BUT supersede detected without ## Delta-spec	1	⚠️ PARTIAL
Artifact is orphan (0 links)	0	❌ BLOCKER

Record the link count and whether delta-spec is present.

Step 8 — S13 Forgeplan check (automation layer)

Run:

validation = forgeplan_validate(id=<target>)
score = forgeplan_score(id=<target>)

Parse:

• validation.status — PASS / FAIL / WARNINGS
• score.r_eff — effective readiness score (float 0.0–1.0)

Scoring:

Condition	Score	Status
validate PASS + r_eff > 0	2	✅ SATISFIED
validate PASS + r_eff = 0	1	⚠️ PARTIAL (likely forgeplan#325 leaf-EVID gap)
validate FAIL	0	❌ BLOCKER

Note: r_eff=0 on a leaf EVID is a known cosmetic issue (forgeplan#325). Flag as PARTIAL, not BLOCKER, unless the artifact is a PRD/RFC/ADR (not EVID itself) and r_eff=0 — that indicates missing child evidence and is a genuine gap.

Step 9 — C4 check (conditional — architectural decisions only)

Apply when: kind is adr or rfc AND the body contains ≥3 distinct module references.

Module detection heuristic — scan body for patterns:

• ### Module (heading prefix)
• **Module <name>** bold pattern
• module X / service X / component X (case-insensitive, where X is a proper noun)
• Named items in a ## Components or ## Modules or ## Services section

If ≥3 distinct module/service/component names are found → C4 check applies.

C4 file detection:

1. Look for a relative link to docs/c4/ in the body (e.g. docs/c4/ADR-NNN.md)
2. Check body for C4 diagram markers: ```mermaid block containing flowchart or C4Context
3. Look for ## C4 Diagrams section or reference to /c4-diagram skill

Scoring:

Condition	Score	Status
C4 file referenced OR mermaid C4 block present in body	2	✅ SATISFIED
C4 mentioned in prose but no file/diagram found	1	⚠️ PARTIAL
≥3 modules detected but no C4 mention at all	0	❌ MISSING
<3 modules detected	N/A	—

Step 10 — RIPER check (production-bug discipline)

Apply when: the target artifact (or its parent PRD/EVID) is a bug-fix artifact for a non-trivial production bug, AND depth is non-tactical. Detect via any of these signals in the artifact title, body, or parent PRD body:

• Title or body mentions: production bug, prod bug, incident, postmortem, post-mortem, outage, regression, race condition, SEV-1, SEV-2, P0, P1, продакшн баг, прод баг, гонка, регрессия, инцидент.
• Parent PRD links to an incident / outage / postmortem context.
• Artifact kind=evidence whose linked parent matches the row 4 of smith routing-map (Bug fix — production, non-trivial).

Skip if: artifact is a trivial hotfix (matches typo, off-by-one, broken link, hotfix, опечатка, хотфикс, single-line fix) — RIPER overhead exceeds value. Mark RIPER as N/A in the report.

Scoring: per linked EVID (or the artifact body itself if kind=evidence), inspect for RIPER Research-mode discipline. Two accepted forms — the contract-conformant form (RFC-018, ADR-010 instance #4) and the legacy form (pre-RFC-018 bug-fix EVIDs):

Condition	Score	Status
Contract form (RFC-018): a standalone Research NOTE + a C4 EVID (from a context distinct from the researcher, e.g. artifact-reviewer) carrying a ## Pinned revision section with verdict SUFFICIENT — the non-freezable Research C4+C6. OR legacy form: a dedicated Research-EVID linked OR a substantive ## Research section in the EVID body (≥10 lines with concrete findings: stack trace, repro steps, suspect commits, hypotheses)	2	✅ SATISFIED
A ## Pinned revision EVID OR ## Research section is present but stub-quality (≤5 lines, no concrete findings), OR the Research C4 verdict is INSUFFICIENT/BIASED, OR mentioned in prose without a dedicated section	1	⚠️ PARTIAL
Bug-fix EVID with NO Research NOTE + ## Pinned revision C6 AND no ## Research section AND no linked Research-EVID — code change without documented investigation	0	❌ BLOCKER

Note (RFC-018): the gold form is now the contract pattern — a Research NOTE pinned by a ## Pinned revision C6 EVID (the non-freezable Research conditional-freeze). The legacy ## Research section remains accepted (backward-compatible) for bug-fix EVIDs created before RFC-018. A Plan/code artifact whose Research pin is stale (the NOTE was edited after the C4 — hash mismatch) scores PARTIAL: re-pin via a fresh C4 before treating it as SATISFIED.

Action item if score <2:

Run /riper (Research mode) before any more code changes on this bug fix. RIPER discipline forbids implementation without a Research artifact — see plugins/fpl-skills/skills/riper/SKILL.md and smith routing-map row 4 (Production bug, non-trivial). Produce a Research NOTE and have a different context (artifact-reviewer) certify it with a ## Pinned revision C6 EVID (or, for a legacy fix, add a substantive ## Research section). Tag the EVID with the methodology label RIPER (production-bug discipline) once it lands.

Record which EVID (if any) carries the Research artifact, and whether the bug was classified trivial (skip) or non-trivial (apply).

Step 11 — Aggregate report

Compute:

• Total score: sum of all applicable layer scores
• Max possible score: 2 × (number of applicable layers, excluding N/A)
• Percentage: (total / max) × 100, rounded to nearest integer

Determine overall verdict:

Percentage	Verdict
100%	✅ COMPLETE
75-99%	STRONG
50-74%	⚠️ NEEDS WORK
25-49%	❌ WEAK
<25%	❌ FAIL

If ANY layer scores 0 (BLOCKER) → append — BLOCKED to the verdict regardless of percentage.

---

Output report format

## /methodology-check report for <ARTIFACT-ID>

**Title**: <artifact title>
**Status**: <status> | **Depth**: <depth> | **Kind**: <kind>

| Layer | Status | Score | Detail |
|---|---|---:|---|
| S10 FPF (design) | ✅ | 2/2 | EVID-NNN: 3 hypotheses (H1/H2/H3) |
| S11 BMAD (quality gate) | ⚠️ | 1/2 | EVID-MMM exists but ## Findings is empty |
| S12 OpenSpec (structure) | ✅ | 2/2 | 3 links (refines EPIC-001, informs PRD-XXX, based_on ADR-YYY) |
| S13 Forgeplan (automation) | ✅ | 2/2 | validate PASS + r_eff=0.95 |
| C4 (arch extension) | N/A | — | <3 modules detected |
| RIPER (production-bug discipline) | N/A | — | not a bug-fix artifact |

### Overall: 7/8 (87%) — STRONG

### Action items
- ⚠️ **S11**: EVID-MMM `## Findings` is empty — dispatch `agents-pro:artifact-reviewer` with adversarial mandate. See [BMAD discipline](../../../CLAUDE.md#bmad-adversarial-review-discipline).
- ℹ️  **S10**: 3 hypotheses present — satisfactory. Consider a 4th edge-case hypothesis if relevant.

### Activation readiness
- ✅ Ready to activate once S11 gap is resolved.
- Blockers preventing activation: 0 hard blockers (S11 is CONCERNS, not BLOCKER).

Zero-EVID case (no evidence linked at all):

## /methodology-check report for PRD-NNN

**Title**: ...
**Status**: draft | **Depth**: standard | **Kind**: prd

| Layer | Status | Score | Detail |
|---|---|---:|---|
| S10 FPF (design) | ❌ | 0/2 | No EVIDs linked — ADI cycle required |
| S11 BMAD (quality gate) | ❌ | 0/2 | No EVIDs linked — Profile B review required |
| S12 OpenSpec (structure) | ❌ | 0/2 | Orphan — no links to any artifact |
| S13 Forgeplan (automation) | ❌ | 0/2 | validate FAIL — no evidence |
| C4 (arch extension) | N/A | — | Not an architectural decision |
| RIPER (production-bug discipline) | N/A | — | Not a bug-fix artifact |

### Overall: 0/8 (0%) — ❌ FAIL — BLOCKED

### Action items
- ❌ **S10**: Run `forgeplan_reason PRD-NNN` (MCP) or `/fpf-reason` skill (interactive). Create EVID with ≥3 hypotheses + ADI cycle. See [FPF ADI discipline](../../../CLAUDE.md#fpf-adi-discipline).
- ❌ **S11**: After S10, dispatch `agents-pro:artifact-reviewer` for adversarial review. EVID body MUST contain `## Findings` with ≥1 item. See [BMAD discipline](../../../CLAUDE.md#bmad-adversarial-review-discipline).
- ❌ **S12**: Link this artifact to its parent Epic or context PRD via `forgeplan_link`. Use relation `refines`, `informs`, or `based_on`.
- ❌ **S13**: After evidence linked, run `forgeplan_validate PRD-NNN` — must return PASS before activating.

### Activation readiness
- ❌ NOT ready to activate — 4 blockers found.

---

Hard rules

1. Never auto-fix layer violations — this skill reports only. The user or orchestrator decides what dispatch to make based on action items.
2. Read-only on the target artifact — do NOT call forgeplan_update, forgeplan_link, forgeplan_new, or any mutation tool.
3. Tactical artifacts get reduced layer check — depth=tactical only checks S12+S13. S10, S11, and RIPER are marked N/A (not a gap, not scored).
4. C4 check is heuristic — false positives (detecting "service" in a non-architectural context) are possible. If the C4 heuristic fires unexpectedly, note it in the report and let the user confirm.
5. RIPER check is heuristic — the bug-fix signal is keyword-based (production / incident / SEV-1/2 / race condition / regression). If the heuristic fires unexpectedly (e.g. the word "regression" appearing in a statistical context, not a bug context), note the suspect signal in the report and let the user override to N/A.
6. r_eff=0 on leaf EVID is cosmetic — do not flag it as a hard BLOCKER for the parent PRD. Note the forgeplan#325 upstream issue.
7. One invocation per artifact — do not recursively check linked artifacts (e.g. do not run methodology-check on the linked EVIDs). This is a flat one-artifact report.

---

Error handling

Error	Response
forgeplan_get not found	Stop. Report ERROR: <ID> not found.
forgeplan_validate unavailable	Mark S13 as ⚠️ PARTIAL with note «forgeplan_validate unavailable».
forgeplan_score unavailable	Mark r_eff as unknown; degrade S13 score to 1 if validate PASS.
No links field in artifact	Fall back to scanning body for EVID-NNN, PRD-NNN patterns.
EVID body fetch fails	Mark that EVID as «body unavailable»; use what remains.

---

Companion skills and references

• /supersede — use after S12 gap: creates delta-spec-compliant successor.
• /decay-watch — scan all active ADRs for fired revisit triggers.
• /decision — record a new decision; run /methodology-check after.
• agents-pro:artifact-reviewer — dispatch for S11 BMAD adversarial review.
• forgeplan_reason <ID> — generate ADI hypothesis cycle (S10 FPF).
• /riper — RIPER orchestrator (Research → Innovate → Plan → Execute → Review); dispatch after a RIPER score <2 on a production-bug artifact.
• EPIC-001 — canonical 4-layer pipeline definition.
• PRD-066 — RIPER auto-routing for production bugs (Step 10 of this skill).
• CLAUDE.md ## FPF ADI discipline, ## BMAD adversarial review discipline, ## OpenSpec delta-spec discipline, ## C4 diagrams for ≥3-module architectural decisions — detailed rules per layer.

---

Anti-patterns

• ❌ Running /methodology-check then auto-dispatching fixes — the skill surfaces gaps; a human or orchestrator with ADI gate decides what to do. Blind auto-dispatch wastes agent turns on gaps that may be intentionally deferred.
• ❌ Using score as a merge gate — methodology-check is advisory. Guardian does the formal gate (guardian reads EVID directly; it doesn't call this skill).
• ❌ Reporting score=0 for tactical artifacts on S10/S11 — tactical artifacts don't require ADI or adversarial review. Correct behaviour: N/A, not BLOCKER.
• ❌ Calling forgeplan_validate without handling WARNINGS gracefully — WARNINGS from validate are not FAIL. Map them to S13=1 (PARTIAL), not S13=0 (BLOCKER).
• ❌ Counting the same hypothesis twice — H1 mentioned in EVID-001 and re-quoted in EVID-002 counts as 1, not 2. De-duplicate by content, not by occurrence.