fix: comprehensive swarm self-audit — 15 files, 30+ fixes across all scales

SS-100 swarm audit (36 agents, 5 domains) + SS-50 verification pass
identified and fixed 30+ issues across the entire repo.

SKILL.md:
- Consensus formula clamped to [0.0, 1.0]
- Sealed criteria count scale-aware (SS-50=6, SS-100=8, SS-250=10)
- Scale-conditional Squad Lead logic in Phase 3
- Depth Guard updated for SS-50/100 flat hierarchy (depth 2)
- Circuit breaker: phase-specific behavior, scale-adjusted thresholds
- Recovery levels L1-L5 fully defined
- Hardening regression policy added
- JSON schema validation recovery path
- Phase sequencing rule corrected
- Phase 7/8 templates scale-aware
- SS-50 agent count ~36-52, SS-100 all 5 domains
- Context capsule timeout/max_depth scale-aware
- Model names canonicalized, Phase 8 banner consistent

Cross-repo (14 files synced with SKILL.md):
config.yml, README.md, CONTRIBUTING.md, agents/, docs/scaling.md,
docs/shadow-scoring.md, docs/use-cases.md, docs/architecture-diagrams.md,
docs/architecture.md, protocols/circuit-breaker.md,
protocols/context-capsule.md, protocols/depth-guard.md,
templates/commander.md, site/src/app/page.tsx, .github/skills/

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: Gregg Cochran

.github/skills/swarm-command/SKILL.md

@@ -30,8 +30,8 @@ Before opening a PR, ensure:
 - [ ] Both SKILL.md copies are identical:
   - `skills/swarm-command/SKILL.md`
   - `.github/skills/swarm-command/SKILL.md`
-- [ ] Agent counts match across all files (SS-50: ~52, SS-100: ~89, SS-250: ~316)
-- [ ] Agent counts verified via `grep -r "~52\|~89\|~316" .`
+- [ ] Agent counts match across all files (SS-50: ~36-52, SS-100: ~89, SS-250: ~316)
+- [ ] Agent counts verified via `grep -r "~36-52\|~89\|~316" .`
 - [ ] Shadow scoring references use Shadow Score Spec format (no separate shadow validator agents)
 - [ ] docs/example-output.md reflects current output format (if output format changed)
 

CONTRIBUTING.md

@@ -136,7 +136,7 @@ If the canary fails, the full pod never deploys. One cheap test prevents many ex
 
 | Scale | Agents | Typical Cost | Hard Cap | Wall-Clock |
 |---|---|---|---|---|
-| **SS-50** | ~52 | $2.50 | $5 | ~30s |
+| **SS-50** | ~36-52 | $2.50 | $5 | ~30s |
 | **SS-100** | ~89 | $5.50 | $10 | ~45s |
 | **SS-250** | ~316 | $10 | $20 | ~65–90s |
 
@@ -249,7 +249,7 @@ T+0s     T+2s       T+5s         T+12s       T+45s      T+65s    T+80s   T+90s
 
 | Scale | Agents | Commanders | Workers | Reviewers | Best For | Wall-Clock |
 |---|---|---|---|---|---|---|
-| **SS-50** | ~52 | 3 | 45 | 3 | Fast bounded tasks | ~30s |
+| **SS-50** | ~36-52 | 2-3 | 30-45 | 3 | Fast bounded tasks | ~30s |
 | **SS-100** | ~89 | 5 | 75 | 8 | Multi-file features and reviews | ~45s |
 | **SS-250** | ~316 | 5 | 250 | 10 | Repo-wide or high-stakes work | ~65–90s |
 
@@ -536,9 +536,9 @@ shadow_scoring:
   enabled: true
   spec_version: "1.0.0"
   conformance_level: "L2"
-  sealed_criteria_count: 10
+  sealed_criteria_count: 10  # max; per-scale: SS-50=6, SS-100=8, SS-250=10
   hardening:
-    enabled: true
+    enabled: true  # SS-50 overrides to disabled
     threshold: 15
 ```
 
@@ -552,7 +552,7 @@ See [docs/scaling.md](docs/scaling.md) for full scaling configuration and cost e
 |---|---|
 | **Nexus** | claude-opus-4.6 |
 | **Commanders** (pool: 9) | claude-opus-4.6, claude-opus-4.5, claude-opus-4.6-1m, claude-sonnet-4.6, claude-sonnet-4.5, claude-sonnet-4, gpt-5.4, gpt-5.2, gpt-5.1 |
-| **Squad Leads** | claude-haiku-4.5, gpt-5.4-mini |
+| **Squad Leads** (SS-250 only) | claude-haiku-4.5, gpt-5.4-mini |
 | **Workers** (pool: 6) | claude-haiku-4.5, gpt-5.4-mini, gpt-5-mini, gpt-4.1, gpt-5.3-codex, gpt-5.2-codex |
 | **Reviewers** (7 pairs) | claude-opus-4.6↔gpt-5.4, claude-opus-4.5↔gpt-5.2, claude-opus-4.6-1m↔gpt-5.1, claude-sonnet-4.6↔gpt-5.3-codex, claude-sonnet-4.5↔gpt-5.2-codex, claude-sonnet-4↔gpt-5.4-mini, claude-haiku-4.5↔gpt-5-mini |
 

README.md

@@ -14,7 +14,7 @@ You are **Swarm Command** 🐝 — a multi-model consensus swarm orchestrator ru
 
 **Personality:** Calm, authoritative swarm commander. Military precision meets collective intelligence. Efficient status updates, clear phase transitions, structured output. You are the Nexus — the brain of the hive.
 
-**⚠️ MANDATORY: Execute ALL phases in sequence. NEVER skip phases.**
+**⚠️ MANDATORY: Execute ALL phases 0-8 in sequence. Phase 5 may overlap with Phase 4. If the circuit breaker trips, proceed to Phase 6 with available bundles, then Phase 7 for partial synthesis.**
 
 **🎭 OUTPUT RULE:** Your visible output is the MISSION BRIEFING and RESULTS. Show phase banners, progress tables, and the final synthesized report. Do not narrate your internal process.
 

agents/swarm-command.agent.md

@@ -2,14 +2,14 @@ swarm_command:
   default_scale: "ss-100"
   scales:
     ss-50:
-      commanders: 3
+      commanders: 3  # max; runtime selects 2-3 most relevant domains
       workers_per_commander: 15  # No squad leads — commanders spawn workers directly
       max_depth: 2               # Nexus[0] → Commander[1] → Worker[2]
       reviewers: 3
       sealed_criteria_count: 6
       cost_ceiling_usd: 5.00
       timeout_s: 60
-      total: ~52
+      total: "~36-52"  # 2 commanders = 36, 3 commanders = 52
     ss-100:
       commanders: 5
       workers_per_commander: 15  # No squad leads — commanders spawn workers directly
@@ -41,7 +41,7 @@ swarm_command:
       conflict_penalty_cap: 0.30
 
   depth_guard:
-    max_spawn_depth: 3
+    max_spawn_depth: 3  # SS-250 max; SS-50/SS-100 use max_depth=2 (no Squad Leads)
     max_workers_per_squad_lead: 5
     worker_agent_types: ["explore", "task"]
     commander_agent_type: "general-purpose"
@@ -121,11 +121,14 @@ swarm_command:
     enabled: true
     spec_version: "1.0.0"
     conformance_level: "L2"
-    sealed_criteria_count: 10  # criteria per task
+    sealed_criteria_count: 10  # max criteria; per-scale overrides in scales section (6/8/10)
     hardening:
-      enabled: true
+      enabled: true  # global default; SS-50 overrides to disabled
       max_cycles: 1
       threshold: 15  # hardening triggers if score > 15%
+      scale_overrides:
+        ss-50:
+          enabled: false  # SS-50: score computed but no fix cycle
     categories:
       - happy_path
       - edge_case

config.yml

@@ -238,7 +238,7 @@ flowchart TD
 ```mermaid
 graph LR
     subgraph SS50["🐝 SS-50"]
-        A50["3 Commanders<br/>15 Workers each<br/>3 Reviewers<br/>~52 agents"]
+        A50["2-3 Commanders<br/>15 Workers each<br/>3 Reviewers<br/>~36-52 agents"]
     end
 
     subgraph SS100["🐝🐝 SS-100"]

docs/architecture-diagrams.md

@@ -39,7 +39,7 @@ Is the task bounded to 1–2 files or one very narrow question?
 
 | Scale | Total Agents | Commanders | Squad Leads | Workers | Reviewers | Best For | Wall-Clock |
 |---|---|---|---|---|---|---|---|
-| **SS-50** | ~52 | 3 | — | 45 | 3 | Fast bounded tasks | ~30s |
+| **SS-50** | ~36-52 | 2-3 | — | 30-45 | 3 | Fast bounded tasks | ~30s |
 | **SS-100** | ~89 | 5 | — | 75 | 8 | Default for real software work | ~45s |
 | **SS-250** | ~316 | 5 | 50 | 250 | 10 | Repo-wide or maximum-confidence work | ~65–90s |
 
@@ -55,11 +55,11 @@ Default: **SS-100**. Use `swarm command ss-250` for full deployment or `swarm co
 
 ```text
 L0: 1 Nexus (claude-opus-4.6)
-L1: 3 Commanders (commander pool — 9 models)
-L2: 45 Workers (worker pool — 6 models)  — 15 per commander, spawned directly
+L1: 2-3 Commanders (commander pool — 10 models)
+L2: 30-45 Workers (worker pool — 6 models)  — 15 per commander, spawned directly
     3 Reviewers (cross-family pairs, spawned by Nexus)
 ──────────────────────────
-Total: ~52 agents
+Total: ~36-52 agents
 Cost:  $1.50 – $3.50
 Time:  ~30s wall-clock
 ```
@@ -68,7 +68,7 @@ Time:  ~30s wall-clock
 
 | Parameter | Value |
 |---|---|
-| Commanders | 3 |
+| Commanders | 2-3 |
 | Domains covered | 2–3 of 5 (auto-selected by task type) |
 | Squad Leads per Commander | — |
 | Workers per Commander | 15 |
@@ -93,7 +93,7 @@ Time:  ~30s wall-clock
 
 ```text
 L0: 1 Nexus (claude-opus-4.6)
-L1: 5 Commanders (commander pool — 9 models)
+L1: 5 Commanders (commander pool — 10 models)
 L2: 75 Workers (worker pool — 6 models)  — 15 per commander, spawned directly
     8 Reviewers (cross-family pairs, spawned by Nexus)
     Shadow Scoring (Nexus-internal, sealed criteria)
@@ -108,10 +108,10 @@ Time:  ~45s wall-clock
 | Parameter | Value |
 |---|---|
 | Commanders | 5 |
-| Domains covered | 3 of 5 (auto-selected by task type) |
+| Domains covered | All 5 |
 | Squad Leads per Commander | — |
 | Workers per Commander | 15 |
-| Reviewers | 8 reviewers (7 cross-family pairs) |
+| Reviewers | 8 reviewers (3-4 cross-family review pairs) |
 | Shadow scoring | 8 sealed criteria, hardening at >15% |
 | Cost ceiling | $10.00 |
 | Timeout cascade | 75/50/35/25s |
@@ -132,10 +132,10 @@ Time:  ~45s wall-clock
 
 ```text
 L0: 1 Nexus (claude-opus-4.6)
-L1: 5 Commanders (commander pool — 9 models)
+L1: 5 Commanders (commander pool — 10 models)
 L2: 50 Squad Leads (claude-haiku-4.5 | gpt-5.4-mini)  — 10 per commander
 L3: 250 Workers (worker pool — 6 models)               — 5 per squad lead
-L4: 10 Reviewers (7 cross-family pairs)
+L4: 10 Reviewers (7 cross-family pairs, cycled to fill 10 slots)
     Shadow Scoring (Nexus-internal, sealed criteria)
 ──────────────────────────
 Total: ~316 agents
@@ -152,7 +152,7 @@ Time:  ~65–90s wall-clock
 | Squad Leads per Commander | 10 |
 | Workers per Squad Lead | 5 |
 | Reviewers | 10 reviewers forming 7 cross-family pairs |
-| Shadow scoring | 10 sealed criteria, hardening at >15% |
+| Shadow scoring | 10 sealed criteria (SS-50: 6, SS-100: 8), hardening at >15% (SS-50: disabled) |
 | Cost ceiling | $20.00 |
 | Timeout cascade | 90/60/40/30s |
 

docs/scaling.md

@@ -98,7 +98,7 @@ After commanders complete and cross-review finishes:
 4. **Classify** using the interpretation scale
 5. **Produce a Gap Report** for each bundle
 
-### Phase 4: HARDENING (Swarm Command Phase 6, continued)
+### HARDENING (Swarm Command Phase 6, continued)
 
 If Shadow Score > 15% for any bundle:
 
@@ -217,9 +217,9 @@ shadow_scoring:
   enabled: true
   spec_version: "1.0.0"
   conformance_level: "L2"
-  sealed_criteria_count: 10
+  sealed_criteria_count: 10  # max; per-scale: SS-50=6, SS-100=8, SS-250=10
   hardening:
-    enabled: true
+    enabled: true  # SS-50 overrides to disabled
     max_cycles: 1
     threshold: 15
   categories:

docs/shadow-scoring.md

@@ -22,7 +22,7 @@ This guide turns the swarm from an impressive concept into a practical tool. Pic
 
 ## SS-50 — Fast Expert Panels
 
-*~52 agents · 3 commanders · 45 workers · ~30 seconds · $1.50–$3.50*
+*~36-52 agents · 2-3 commanders · 30-45 workers · ~30 seconds · $1.50–$3.50*
 
 ### 1. 🔥 Stack Trace Whisperer
 

docs/use-cases.md

@@ -75,10 +75,11 @@ The HALF-OPEN probe MUST:
 
 | Layer | Agents | Threshold to OPEN | Cooldown | Probe Size |
 |---|---|---|---|---|
-| **Nexus (L0)** | Monitors 5 Commanders | 3/5 commanders fail (60%) | 10s | 1 commander re-dispatch |
-| **Commander (L1)** | Monitors 10 Squad Leads | 5/10 squad leads fail (50%) | 5s | 1 squad lead re-dispatch |
-| **Squad Lead (L2)** | Monitors 5 Workers | 3/5 workers fail (50%) | 3s | 1 canary worker |
-| **Reviewer (L4)** | Monitors review mesh | 3/5 reviews fail (50%) | 5s | 1 review re-dispatch |
+| **Nexus (L0)** | Monitors 2-5 Commanders | SS-50: 2+ fail (≥50%), SS-100/250: 3/5 fail (60%) | 10s | 1 commander re-dispatch |
+| **Commander (L1, SS-250)** | Monitors 10 Squad Leads | 5/10 squad leads fail (50%) | 5s | 1 squad lead re-dispatch |
+| **Commander (L1, SS-50/100)** | Monitors 15 Workers | 8/15 workers fail (50%) | 5s | 1 canary worker |
+| **Squad Lead (L2, SS-250 only)** | Monitors 5 Workers | 3/5 workers fail (50%) | 3s | 1 canary worker |
+| **Reviewer** | Monitors review mesh | 3/5 reviews fail (50%) | 5s | 1 review re-dispatch |
 
 ### Failure Definitions