council

name	council
description	Run 8-persona Council validation on a feature/ADR, then auto-create a Linear ticket if score >= 8/10.
argument-hint	<description of what to build>

Council Validation — $ARGUMENTS

Three-Stage Council

This skill drives Stage 1 (ticket pertinence) and Stage 2 (plan validation) — both prose-based, 8-persona scoring per the canonical Team Coca jury (HEG-PAT-003).

Stage 3 (code review) runs after implementation on the actual diff, via scripts/council-review.sh (4-axis: conformance, debt, attack_surface, contract_impact). It is not invoked by this skill — it is wired to the pre-push hook (CAB-2048) and council-gate.yml CI workflow (CAB-2049). See .claude/docs/council-s3.md for full docs.

S1 (this skill, prose)  →  S2 (this skill, plan)  →  implementation  →  S3 (council-review.sh, diff)  →  merge

Step 1: Gather Context

Read and understand the feature/ADR/change description. Gather context from:

• memory.md — current sprint, related work
• plan.md — where this fits in priorities
• Relevant code files if technical

Step 1b: Context Compiler — Impact Score

Run the Context Compiler to get the blast radius before scoring:

1. Determine the primary component from $ARGUMENTS (or from the ticket description if CAB-XXXX is provided)
2. Run: docs/scripts/build-context.sh --component <component> --intent "$ARGUMENTS"
   • If a ticket ID is available: add --ticket "CAB-XXXX"
3. Read the generated context pack in docs/context-packs/
4. Extract the Impact Score and level (LOW/MEDIUM/HIGH/CRITICAL)
5. Include the Impact Score in each persona's evaluation context
6. If CRITICAL (>= 31): flag it explicitly in the Council report header

If build-context.sh or stoa-impact.db is not available, skip this step with a warning and proceed without impact scoring.

Step 2: Run the Council — 8 Personas

Evaluate $ARGUMENTS through each of the 8 canonical personas. All 8 are mandatory — no subset selection, no quick mode. The canonical roster is defined in hegemon/patterns/validation/HEG-PAT-003-team-coca-adversarial-validation.md.

Domain N/A baseline (calibration rule from CAB-2054 Phase 4): when a ticket does not touch a persona's domain (e.g., a backend infra ticket has no webapp surface for N3m0), score the persona at 8/10 (neutral-positive: "no concern detected"), not 6-7. Lower scores must be backed by an actual identified concern. This rule keeps the 8-persona average stable vs the historical 4-persona baseline (calibration drift < ±0.3 on 4/5 historic tickets).

1. Chucky (Devil's Advocate) — Score /10

• Challenges assumptions and finds weaknesses
• Questions: Is this really needed? What could go wrong? What's the hidden complexity? How would I break this?
• Red flags: scope creep, over-engineering, unclear requirements, missing failure modes
• Focus: risk assessment, edge cases, failure modes, penetration testing mindset

2. N3m0 (Webapp Security) — Score /10

• Evaluates frontend exposure, UX, accessibility, developer experience
• Questions: What UI surfaces are exposed? Is the keyboard flow valid? Are there a11y blockers? Do error messages leak info? Is the DX of the change reasonable?
• Red flags: missing ARIA, keyboard traps, info-leak in error responses, UI surface that bypasses backend auth, fragile selectors
• Focus: web application security, accessibility (WCAG), UX safety, frontend attack surface

3. Gh0st (Supply Chain) — Score /10

• Evaluates dependencies, SBOM, lockfile integrity, transitive trust
• Questions: What new deps does this pull? Are they pinned? Lockfile drift? Any transitive CVEs? Is the source repo trustworthy?
• Red flags: unpinned versions, dep from unknown publisher, lockfile not updated, vendored binary blobs, npm/pip postinstall scripts
• Focus: supply chain security, dependency trust, SBOM hygiene

4. Pr1nc3ss (Social Engineering) — Score /10

• Evaluates human factors, trust boundaries, phishing surface
• Questions: How would someone trick a user into bypassing this? What does the failure path teach an attacker? Are error messages too informative? Trust prompts clear?
• Red flags: error messages that enumerate users / disclose internals, ambiguous consent UI, trust dialogs that habituate clicks, copy-paste-prone secret flows
• Focus: human factor in security, trust boundary design, anti-phishing posture

5. OSS Killer (VC Skeptique) — Score /10

• Evaluates market viability, competitive moat, user value
• Questions: Does this differentiate STOA? Would a user pay for this? Is this a feature or a product? What's the ROI?
• Red flags: me-too features, no measurable impact, building for nobody, parity work without margin uplift
• Focus: competitive advantage, adoption potential, ROI, business viability

6. Archi 50x50 (Architecte Veteran) — Score /10

• Evaluates architectural quality, technical debt, scalability
• Questions: Does this follow existing patterns? Is the complexity justified? Will this age well? Does it work at enterprise scale?
• Red flags: wrong abstraction level, coupling, breaking existing contracts, hidden statefulness
• Focus: technical excellence, maintainability, performance, long-horizon evolution

7. Better Call Saul (Legal/IP) — Score /10

• Evaluates legal risks, compliance, IP protection, content safety
• Questions: Any license issues? GDPR/DORA compliance? Competitive claims risks? Customer data exposure?
• Red flags: hardcoded secrets, unverified competitive claims, missing disclaimers, GPL contamination, PII leakage
• Focus: legal safety, regulatory alignment, IP protection, compliance posture

8. Gekk0 (Monétisation / GTM) — Score /10

• Evaluates revenue alignment, pricing tier fit, time-to-revenue, GTM messaging
• Named after Gordon Gekko ("Wall Street") — ruthless monetization lens. Internal alias.
• Questions: Can we charge for this? Which pricing tier (Open Source / Enterprise / Managed)? Does it shorten the sales cycle? Can sales explain the value prop in 30 seconds?
• Red flags: cool feature with no revenue path, cannibalizes higher-tier offering, sales team can't explain it, misaligned with commercial positioning
• Focus: revenue alignment, pricing tier fit, time-to-revenue, sales enablement, GTM messaging

Present each persona's evaluation:

## Council Validation

### Chucky (Devil's Advocate) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences: key concerns or approval>
**Adjustments**: <specific changes required, or "None">

### N3m0 (Webapp Security) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

### Gh0st (Supply Chain) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

### Pr1nc3ss (Social Engineering) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

### OSS Killer (VC Skeptique) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

### Archi 50x50 (Architecte Veteran) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

### Better Call Saul (Legal/IP) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

### Gekk0 (Monétisation / GTM) — X/10
**Verdict**: Go | Fix | Redo
<2-3 sentences>
**Adjustments**: <specific changes required, or "None">

---
**Persona Average**: X.XX/10 (averaged over 8 personas)
**Impact Score**: N (LEVEL) | Modifier: -Y.Y
**Final Score**: X.XX/10
**Global Verdict**: Go | Fix | Redo
**Adjustments to apply**: <numbered list of all adjustments from all 8 personas>

Step 3: Decision Gate

Impact Score Modifier

The Impact Score from Step 1b applies a risk modifier to the persona average:

Impact Level	Score Range	Modifier	Rationale
LOW	0-5	0	Isolated change, no risk adjustment
MEDIUM	6-15	0	Manageable blast radius
HIGH	16-30	-0.5	Significant cross-component risk
CRITICAL	31+	-1.0	Major blast radius, forces careful review

Final Score = Persona Average + Impact Modifier

If Step 1b was skipped (no DB), Impact Modifier = 0.

Verdict Thresholds (applied to Final Score)

Final Score	Action
>= 8.0	Go — Auto-create Linear ticket (Step 4)
6.0 - 7.9	Fix — List adjustments, ask user to confirm, then create ticket
< 6.0	Redo — Fundamental issues, do NOT create ticket. Propose alternatives.

Step 3b: Apply Stage 1 Label on Linear

After scoring, apply the appropriate Stage 1 label to the Linear ticket (if one exists or is being created):

Score Range	Label	Color
>= 8.0	council:ticket-go	green (#0e8a16)
6.0 - 7.9	council:ticket-fix	amber (#e4b400)
< 6.0	council:ticket-redo	red (#d73a49)

Apply via: linear.update_issue(id, labels: [existing_labels..., "council:ticket-go|fix|redo"])

If the ticket already has a council:ticket-* label, replace it (remove old, add new).

Step 4: Auto-Create Linear Ticket

If Council passes (Go or Fix-then-confirmed), create the ticket using Linear MCP:

linear.create_issue(
  title: "<type>(scope): <short description>",
  description: <see template below>,
  team: "<LINEAR_TEAM_ID>",
  project: "<LINEAR_PROJECT_ID>",
  assignee: "<LINEAR_ASSIGNEE_ID>",
  estimate: <fibonacci points>,
  priority: <1=Urgent, 2=High, 3=Normal, 4=Low>,
  labels: [<type label>, <priority label>, "hlfh:validated", "council:ticket-go|fix", "instance:<detected-instance>"],
  state: "Todo"
)

Issue Description Template

## Context
<1-2 paragraphs: what and why>

## Council Validation — X.XX/10 {Go|Fix} | Impact: N (LEVEL, modifier -Y.Y)

| Persona | Score | Verdict |
|---------|-------|---------|
| Chucky (Devil's Advocate) | X/10 | Go |
| N3m0 (Webapp Security) | X/10 | Go |
| Gh0st (Supply Chain) | X/10 | Go |
| Pr1nc3ss (Social Engineering) | X/10 | Go |
| OSS Killer (VC Skeptique) | X/10 | Go |
| Archi 50x50 (Architecte Veteran) | X/10 | Go |
| Better Call Saul (Legal/IP) | X/10 | Go |
| Gekk0 (Monétisation / GTM) | X/10 | Go |

### Adjustments Applied
1. <adjustment 1>
2. <adjustment 2>

## Scope
<Bullet list of what's in/out of scope>

## Implementation Phases
<Numbered list of phases with estimated LOC>

## DoD
- [ ] <acceptance criteria 1>
- [ ] <acceptance criteria 2>
- [ ] State files updated (memory.md, plan.md)
- [ ] CI green

Estimate Guide

Complexity	Points	Examples
Trivial	1-2	Config change, docs update, single-file fix
Small	3-5	Single component, <150 LOC, clear scope
Medium	8-13	Multi-component, <300 LOC, needs design
Large	21-34	Cross-cutting, >300 LOC, multiple PRs
Epic	55+	Multi-sprint, architectural change

Priority Mapping

Signal	Priority	Linear Label
Demo blocker, production down	1 (Urgent)	P0-Urgent
This week, user-facing	2 (High)	P1-High
This sprint	3 (Normal)	P2-Medium
Backlog, nice-to-have	4 (Low)	P3-Low

Step 4b: Stage 2 — Plan Validation

After ticket creation and before implementation begins, run Stage 2 to validate the implementation plan.

When Stage 2 Runs

Level	Stage 2?	Trigger
L1 Issue-to-PR	Yes	/go triggers Stage 2, /go-plan triggers implement
L3 Linear Dispatch	Yes	/go triggers Stage 2, /go-plan triggers implement
L3.5 Autopilot	No (batch, Stage 1 only)	Unchanged
L5 Multi-Agent	No (batch, Stage 1 only)	Unchanged
/council local	Yes	/council plan "..." triggers Stage 2

Stage 2 Rubric (same 8 personas as Stage 1, plan-focused questions)

1. Chucky (Devil's Advocate) — Score /10

• Risks documented? Edge cases covered by tests?
• Failure modes identified? Rollback plan?
• Hidden dependencies that could block?

2. N3m0 (Webapp Security) — Score /10

• Does the UI plan address a11y (keyboard, screen reader, ARIA roles)?
• Are exposed UI surfaces (new routes, public endpoints) enumerated?
• Do error responses leak internal info? Are user-input boundaries validated?
• Is the DX of the change reasonable for downstream consumers?

3. Gh0st (Supply Chain) — Score /10

• Are new deps pinned with integrity hashes? Lockfile updated?
• Any dep from an untrusted publisher? Transitive CVE check?
• Vendored binaries? Postinstall scripts?
• SBOM regeneration documented?

4. Pr1nc3ss (Social Engineering) — Score /10

• Do error messages enumerate users / disclose internals?
• Trust prompts / consent UI clear and unambiguous?
• Any copy-paste-prone secret flows? Habituated trust dialogs?
• Anti-phishing posture preserved?

5. OSS Killer (VC Skeptique) — Score /10

• LOC justified? Delivers the announced value?
• ROI of dev effort vs. user impact?
• Could this be done simpler? Faster alternative explored?

6. Archi 50x50 (Architecte Veteran) — Score /10

• Right level of abstraction? Existing patterns respected?
• Breaking changes documented? Migration path clear?
• Test strategy adequate? Statefulness explicit?

7. Better Call Saul (Legal/IP) — Score /10

• No secrets hardcoded in the plan? GDPR/DORA impacts noted?
• License implications of new dependencies (incl. transitive)?
• Content safety of any new user-facing text?
• Customer data exposure path?

8. Gekk0 (Monétisation / GTM) — Score /10

• Does the plan preserve the monetization hook? Pricing tier boundary intact?
• Breaking change to commercial offering? Cannibalization risk?
• Post-merge marketing comms needed? Sales team enabled?
• Time-to-revenue acceleration or regression?

Stage 2 Labels

Score Range	Label	Color
>= 8.0	council:plan-go	teal (#006b75)
6.0 - 7.9	council:plan-fix	dark amber (#b45309)
< 6.0	council:plan-redo	dark red (#8b0000)

Apply via: linear.update_issue(id, labels: [existing_labels..., "council:plan-go|fix|redo"])

Stage 2 Flow

Stage 1 passes (council:ticket-go)
  ↓ /go
Claude writes implementation plan (posted as issue comment)
  ↓ plan posted
Stage 2 validates the plan (4 personas, plan-focused rubric)
  ↓ council:plan-go applied to Linear
Comment `/go-plan` to approve and start implementation

Step 5: Update plan.md

After ticket creation, append to the appropriate section in plan.md:

- [ ] CAB-XXXX: <title> (<points> pts) — Council X.XX/10

Step 6: Report to User

Council: X.XX/10 — Go
Ticket: CAB-XXXX (<points> pts)
Linear: <URL>
plan.md: updated
Next: say "go" to start implementation, or "adjust <feedback>" to revise

Step 4c: Stage 3 — Automated Code Review (post-implementation)

Stage 3 is not run by this skill. It runs automatically against the diff once implementation is complete, via scripts/council-review.sh. Shape:

Axis	Role
conformance	Repo conventions, style, naming, structure
debt	Shortcuts, TODOs, test gaps, premature abstractions
attack_surface	Input validation, authn/z, secrets, SSRF, PII, CORS, deps
contract_impact	Cross-component contracts (skipped when docs/stoa-impact.db is stale)

Exit codes: 0 APPROVED (avg >= 8.0), 1 REWORK, 2 technical failure.

Trigger points (when wired):

Level	Trigger	How
Local pre-push	git push	Pre-push hook extension (CAB-2048)
CI	PR opened	.github/workflows/council-gate.yml feature-flagged via vars.COUNCIL_S3_ENABLED (CAB-2049)
Manual	Any diff	scripts/council-review.sh --ticket CAB-XXXX --diff <range>

When S3 returns REWORK, address the per-axis blockers/warnings and re-run. Full documentation: .claude/docs/council-s3.md.

Rules

• Never skip the Council for features >= 5 pts
• Always include DoD in the Linear ticket
• Adjustments are mandatory — if a persona says Fix, the adjustment must be in the DoD
• One ticket per feature — don't create sub-tasks (those come during implementation)
• Score honestly — don't inflate scores to pass the gate
• For ADRs: use /create-adr skill after Council passes (ADR lives in stoa-docs)
• For mega-tickets (>= 30 pts): add mega-ticket label automatically
• Stage 1 labels (council:ticket-*) are applied after every Council scoring (Step 3b)
• Stage 2 labels (council:plan-*) are applied after plan validation (Step 4b)