tigris-panel

name: tigris-panel description: Run a TIGRIS Parliament review — the three-phase adversarial procedure (Stakes / Argument / Amendment) that produces a review-as-game-record for a board game or a TIGRIS-internal artifact. Invoked after a game's DESIGN is complete or a spec is ready for scrutiny. Outputs: tierA-stakes.md, playtests/PT*-argument.md, panel/SUMMARY.md, amendments.md.

tigris-panel — Parliament procedure

60-second quickstart

You:    "Run tigris-panel on games/0001-parliament/"
Claude: [reads design.md, axis-pool.md, personas/, forbidden-words.md]
        [dispatches 7 designer personas to DRAFT 3 axes each in snake order]
        [writes games/0001-parliament/tierA-stakes.md]
        [reports: "21 stakes drafted, GATE passed — 3 collision candidates identified"]
        [runs ARGUMENT phase: narrated playthrough with attacks/defends/collisions]
        [writes games/0001-parliament/playtests/PT01-argument.md]
        [runs AMENDMENT phase deterministically]
        [writes games/0001-parliament/amendments.md + updates axis-pool.md]
        [writes panel/SUMMARY.md]
        [reports: "review complete — 5 earned, 2 refuted, 1 collision, 1 axis adopted"]

First invocation of the session, Claude should also announce "using tigris-panel to run the Parliament review on " before reading anything.

What this skill does

Executes the three-phase Parliament procedure defined in docs/specs/2026-04-19-tigris-v2.0-design.md §8.3–8.6. Produces an argument-record, not a consensus document.

Two modes:

• game-review — artifact is under games/NNNN-<slug>/; full roster (7 designers + optional 5 lenses); output under games/<slug>/.
• spec-review — artifact is under docs/specs/ or personas/ or .claude/skills/; designers only (lenses don't parliament on processes); output under docs/specs/reviews/<artifact-slug>/ with the same Parliament shape.

Preconditions

• Artifact exists and has YAML frontmatter with slug, stage, version, rubric_version: v2.0.
• All 7 designer personas exist at personas/designers/*.md with preferred_axes field populated.
• personas/axis-pool.md exists and has 24 live-or-adopted axes.
• personas/forbidden-words.md exists (vocabulary discipline enforced at write-time).
• For originals: CONCEPT has declared anchor_persona and anchor_axis. For reviews: CONCEPT may declare these or skip (anchor-cell may be left null in spec-review mode).

Procedure

Step 1 — Load context

1. Read the artifact.
2. Read personas/axis-pool.md (current live axes, adopted list, retired list).
3. Read personas/playtest-rubric.md (the Parliament-shape rubric v2.0).
4. Read personas/forbidden-words.md.
5. Read docs/specs/2026-04-19-tigris-v2.0-design.md §8 for procedure details.
6. Determine mode (game-review vs spec-review) from artifact path.
7. Create output directory.

Step 2 — TIER-A: STAKES (draft phase)

1. Determine seating: 7 designer personas (spec-review) or 7 designers + optional 5 lenses (game-review).
2. Establish draft order:
   • Round 1: user-specified order, usually "anchor persona last" (forces them onto shared ground).
   • Round 2: reverse of round 1.
   • Round 3: same as round 1.
3. For each draft slot, dispatch a mini-task:
   • Persona reads their preferred_axes list.
   • Persona reads the Pool's current state (live / adopted / retired).
   • Persona picks one axis not already drafted this game.
   • A persona may pick a retired axis at cost of 2 draft-slots (skip their next draft).
   • Record draft in tierA-stakes.md.
4. After drafts complete, each persona stakes their 3 axes:
   • For each target player count of the design, assign score 0–10 with a rule-grounded one-sentence justification.
   • Record in persona's section of tierA-stakes.md.
5. Publish tierA-stakes.md.

Step 3 — GATE check

Verify all three GATE conditions from personas/playtest-rubric.md:

1. Anchor stake viable? (If declared; otherwise mark N/A for spec-reviews.)
2. Draft coverage ≥ 3 bands of the 4 (A/B/C/D)?
3. ≥ 1 collision candidate identifiable?

If pass: proceed to Step 4. If fail: return to user with specific failure mode and a re-draft proposal.

Step 4 — TIER-B: ARGUMENT (narrated playthrough)

1. For games: seeded-RNG narrated playthrough covering 2p / 3p / 4p serially. For specs: walk the spec section-by-section with personas commenting.
2. At each pre-declared moment (start, turn 1 / §1, turn 5 / §5, mid, late, end):
   • Each persona reviews their stake(s) applicable at this moment.
   • Each persona may declare: hold, attack , defend, or collide-with .
   • Attacks require cited evidence; defenses require cited evidence; collisions require adjacency argument.
3. FORCED-ENGAGEMENT (v2.1) — inserted between the penultimate and final moment-anchor. For each drafted stake with 0 marks at this point, the owning persona must declare one of: HOLD-EXPLICIT (cost 1 token), RETIRE-EXPLICIT (refund 1 token, +0.5 refute mark), or LAST-CALL-STAKE (double tokens on final round). Stakes still at 0-marks after this phase become silent-retire at endgame (−3 points, +1 toward retirement, persona-retirement flag at 3+ in a game).
4. Record every action in playtests/PT<NN>-argument.md with moment-anchors (turn numbers or section numbers).
5. At game end, classify each stake: earned / contested / refuted / hold-explicit / retire-explicit / silent-retire. The ignored state is no longer terminal in v2.1.

Step 5 — TIER-C: AMENDMENT (deterministic pass)

1. Read personas/axis-pool.md Rubric Ledger.
2. For each stake in this game, update ledger row:
   • +1 to earned / contested / refuted / ignored count on that axis.
3. Check adoption triggers (≥ 2 earned across ≥ 2 games):
   • Mark axis adopted.
   • Bump rubric version in personas/playtest-rubric.md (v2.0 → v2.1).
4. Check retirement triggers (≥ 2 refuted across ≥ 2 games):
   • Mark axis retired.
5. Write games/<slug>/amendments.md with the full change record.
6. Log new innovations to personas/playtest-innovations.md (include trigger_pattern field).

Step 6 — Synthesize SUMMARY.md

Write panel/SUMMARY.md with:

---
name: Parliament Summary — <artifact>
slug: SUMMARY
stage: panel
version: 1.0.0
rubric_version: v2.0
bet_version: parliament
artifact_under_review: <path>
personas_seated: [list]
total_stakes: <N>
earned: <N> / refuted: <N> / contested: <N> / ignored: <N>
collisions: <N>
axes_adopted: [list]
axes_retired: [list]
created: <today>
---

Body structure (lean — this is an argument record, not an essay):

1. Headline — one sentence naming the argument's shape.
2. Earned stakes — table: persona, axis, count, score, defending moment.
3. Refuted stakes — table: persona, axis, count, score, refuting moment.
4. Collisions — each collision with both personas, both axes, resolving moment.
5. Ignored stakes — flagged as silent failures with persona names (persona-retirement candidates if repeated).
6. Amendment outcome — axes adopted, axes retired, rubric version change.
7. Success criteria check — did the review meet playtest-rubric.md's 4 success criteria?

Step 7 — Update TRACKER.md

Append a row:

Step 8 — Hand off

Report to user:

• Paths to all output artifacts.
• Success criteria check result.
• Total earned / refuted / ignored / collision counts.
• Axes adopted or retired.
• Next step: HANDOFF (if game) or spec amendment integration (if spec-review).

Invariants

• Stakes are never averaged. Aggregate scores are banned.
• Every attack, defense, and collision cites a specific moment (turn number, section number) and a specific rule or mechanic.
• Ignored stakes are reported, not hidden. They are as meaningful as earned stakes.
• Axis adoption and retirement are deterministic from the ledger — no user approval for the default path.
• Reviews are immutable once written. A revision creates a new version directory.
• Forbidden-words compliance is enforced at stake-write time, not post-hoc.

Mode differences

Example invocation

"Run tigris-panel on games/0001-parliament/"

Expected output:

games/0001-parliament/
├── tierA-stakes.md          # 21 stakes, 3 per persona
├── playtests/
│   └── PT01-argument.md     # argument record with moment-anchors
├── amendments.md            # what this game adopted/retired
├── panel/
│   └── SUMMARY.md           # lean argument record
└── handoff.md               # (written by a later HANDOFF step)

Plus updates to:

• personas/axis-pool.md — ledger updated
• personas/playtest-rubric.md — version bumped if adoption
• personas/playtest-innovations.md — new innovations appended
• TRACKER.md — new row

What this skill does NOT do

• Invoke HANDOFF — that's a separate step.
• Re-score prior games on amendment — forward-only versioning means prior scores are locked.
• Approve or reject the artifact — the output is evidence; user decides.
• Generate physical components or renderings — Parliament's design doc is markdown-only for v2.0.