Propose: compressed Squad coordinator prompt (~49% smaller, behavior-preserving) [DRAFT]

[tamirdresher]

DRAFT / DO-NOT-MERGE. This is a proposal — a compressed source-template
candidate for .github/agents/squad.agent.md, offered for review and side-by-side
trial. Adopting it changes no live coordinator behavior on its own. Merge is gated on
governance review (Picard/Data) and the maintainer's call.

What this is

A behavior-preserving compression of the Squad coordinator source template.
The coordinator prompt is always-on (injected into every Squad session), so its size
is paid on every turn. This PR proposes cutting it ~in half without dropping any
coordinator behavior.

Base & provenance (pinned)

	value
Base	dev @ eaf3ccac
Base file	.github/agents/squad.agent.md — 79,966 B, sha256 1F69CEB0…
Proposed file	40,305 B, sha256 E10036FA…
Reduction	−49.6% bytes, 1080 → 302 lines (~−46% input tokens/call by tiktoken)
Frontmatter	tools: ["*"] + version: 0.0.0-source — byte-identical to dev

Behavior preservation

Compression is prose-only. Every high-risk coordinator path is kept intact
(grep-verified present in the proposed file):

• Routing table + Response-Mode selection
• State-backend handshake HALT semantics
• Reviewer strict-lockout protocol
• Init Mode Phase-1 ask_user confirmation gate
• cross-squad dispatch (product vs. English disambiguation)
• STOP/dispatch gate ("never do domain work inline")
• Coordinator canary marker

The two dev-only sections added since the compression base —
Per-Agent Context Tier and the Shared-worktree guard — are carried forward,
compressed in the same dense style as the sibling Per-Agent Reasoning Effort
paragraph (every resolution layer, config key, ack string, and safety semantic
preserved).

How it was verified

1. The compression was produced from an independently Q-approved shortening of the
   installed coordinator (a frozen SHA), which preserved the high-risk paths
   byte-for-byte.
2. This PR's file = that approved compression + exactly the 5 edits that make it
   match dev's base (tools, version ×2, the two dev-only blocks). Diff vs. the
   approved artifact shows those edits and nothing else.
3. All 7 high-risk anchors + the canary confirmed present; frontmatter byte-identical
   to dev.

Honest scope of the claim

• ✅ Deterministic: −49.6% bytes / −46% input tokens per call (hardware-independent).
• ✅ Mechanism (local, Q_mech): on a local model with cache defeated, the smaller
  prompt prefilled ~2.25× faster cold / ~1.6× faster warm; input-token saving
  holds in every case. This is the mechanism by which prompt size can affect a session.
• ❌ NOT claimed: any real Copilot-CLI end-to-end performance result (Q_host).
  That remains blocked/unproven pending a trustworthy host-telemetry path
  (see Runtime identity beacon for coordinator canary cases 3 & 4 (host-owned pre-ingestion state machine) #1518). No real-session speed/quality claim is made here.

How to review

1. Verify the pinned SHAs above; diff the single file.
2. Run the coordinator behavior-equivalence suite over the high-risk paths
   (routing, state-backend HALT, reviewer lockout, init confirmation, cross-squad
   dispatch). Treat any semantic mismatch as a failed retention claim.
3. If dev has moved, rebase — the delta set vs. dev is small and mechanical
   (frontmatter + the two carried-forward blocks).

Related: #1518 (host telemetry), #1517.

Governance

Pending Picard/Data governance review and maintainer authorization before any
adoption. Filed as draft deliberately.

_{Provenance: proposed file sha256 E10036FA…; base dev@eaf3ccac file sha256 1F69CEB0….}

[github-actions]

🛫 PR Readiness Check

ℹ️ This comment updates on each push. Last checked: commit e2128a8

PR Scope: 🔧 Infrastructure

⚠️ 2 item(s) to address before review

Status	Check	Details
✅	Single commit	1 commit — clean history
❌	Not in draft	PR is still in draft — mark as ready for review when done
✅	Branch up to date	Up to date with dev
❌	Copilot review	No Copilot review yet — it may still be processing
✅	Changeset present	No source files changed — changeset not required
✅	Scope clean	No .squad/ or docs/proposals/ files
✅	No merge conflicts	No merge conflicts
✅	Copilot threads resolved	No Copilot review threads
✅	CI passing	All checks passing

Files Changed (1 file, +175 −959)

File	+/−
.github/agents/squad.agent.md	+175 −959

Total: +175 −959

---

This check runs automatically on every push. Fix any ❌ items and push again.
See CONTRIBUTING.md and PR Requirements for details.

[github-actions]

🟢 Impact Analysis — PR #1529

Risk tier: 🟢 LOW

📊 Summary

Metric	Count
Files changed	1
Files added	0
Files modified	1
Files deleted	0
Modules touched	1

🎯 Risk Factors

• 1 files changed (≤5 → LOW)
• 1 module(s) touched (≤1 → LOW)

📦 Modules Affected

ci-workflows (1 file)

• .github/agents/squad.agent.md

---

This report is generated automatically for every PR. See #733 for details.

[bradygaster]

#1517 has now merged into dev, establishing the new three-state/HEAD-canary baseline. Per the approved triage plan, this draft remains parked until it is rebased onto current dev and an independent reviewer re-verifies the high-risk coordinator behavior anchors against the compressed prompt. No promotion or merge until that governance pass is complete.