peaks-loop 4.0.25 → 4.0.26

package/dist/services/memory/llm-reranker.js

@@ -35,10 +35,9 @@
  *
  * Why 4-bytes-per-token:
  *
- *   - `headroom-client.ts:60` already uses `BYTES_PER_TOKEN = 4` as
- *     its rough English-text approximation. Reusing the same constant
- *     keeps token estimates comparable across the headroom + rerank
- *     pipelines in the AC-ZA-5 benchmark.
+ *   - `BYTES_PER_TOKEN = 4` is the standard rough English-text
+ *     approximation (1 token ≈ 4 bytes). It keeps token estimates
+ *     stable and comparable in the AC-ZA-5 benchmark.
  *
  * Out of scope (YAGNI per Karpathy #2 Simplicity First):
  *
@@ -49,9 +48,8 @@
  *   - No integration with `memory-search-service.ts` (Z-A is a spike;
  *     the wiring into `peaks memory search` is Z-B).
  */
-// Approximate 1 token = 4 bytes for English text. Matches
-// `headroom-client.ts:60` so the AC-ZA-5 benchmark can compare
-// rerank input cost against headroom compressed cost directly.
+// Approximate 1 token = 4 bytes for English text. Standard rough
+// approximation used across the pipeline (AC-ZA-5 benchmark).
 const BYTES_PER_TOKEN = 4;
 // Hard caps so a pathological query / 60-candidate list cannot blow
 // up the LLM prompt. These are conservative defaults; Z-B may tune.
@@ -62,7 +60,7 @@ const MAX_DESCRIPTION_CHARS = 240;
 const DEFAULT_CHAT_TIMEOUT_MS = 5_000;
 /**
  * Estimate token count for a string. `1 token ≈ 4 bytes` for English
- * text — same approximation as `headroom-client.ts:60`.
+ * text (standard rough approximation).
  */
 export function estimateTokens(text) {
     return Math.ceil(Buffer.byteLength(text, 'utf8') / BYTES_PER_TOKEN);

package/dist/services/memory/memory-search-service.d.ts

@@ -59,35 +59,3 @@ export declare function loadMemoryIndex(projectRoot: string): MemoryIndexSnapsho
  * spec §Component Details).
  */
 export declare function searchMemory(input: MemorySearchInput): MemorySearchResult[];
-/**
- * Envelope for the optional headroom-compressed version of the search
- * result text. The compressed text is intended for the LLM-side prompt
- * (it preserves the entry names + descriptions in a more token-efficient
- * form); the structured `matches` array is unchanged and remains the
- * machine-readable form. `null` when compression was not requested,
- * did not run, or failed (see `warning` for the reason).
- */
-export interface CompressedResultsEnvelope {
-    readonly mode: 'balanced' | 'aggressive' | 'conservative';
-    readonly originalSize: number;
-    readonly compressedSize: number;
-    readonly compressionRatio: number;
-    readonly compressedText: string;
-    readonly warning: string | null;
-}
-export interface MemorySearchWithResults {
-    readonly matches: MemorySearchResult[];
-    readonly compressedResults: CompressedResultsEnvelope | null;
-}
-export interface MemorySearchOptions {
-    /** When true, read preferences.headroom.perTouchpoint.memorySearch and
-     *  compress joined match text via headroom-ai. Falls back silently to
-     *  the structured `matches` array when compression is unavailable. */
-    readonly compressResults?: boolean;
-}
-/**
- * Run the search + optionally compress the joined match text via headroom.
- * The structured `matches` array is the canonical output. The
- * `compressedResults` is a derived view for LLM-side prompt assembly.
- */
-export declare function searchMemoryWithResults(input: MemorySearchInput, options?: MemorySearchOptions): Promise<MemorySearchWithResults>;

package/dist/services/memory/memory-search-service.js

@@ -1,9 +1,6 @@
 import { existsSync, readFileSync } from 'node:fs';
 import { join } from 'node:path';
 import { fuzzyMatchWithKey } from '../fuzzy-matching/fuzzy-match-service.js';
-import { compressPrompt } from '../context/headroom-client.js';
-import { loadPreferences } from '../preferences/preferences-service.js';
-import { shouldCompressResults } from '../context/headroom-prefs.js';
 const DEFAULT_LIMIT = 6;
 /**
  * Read `.peaks/memory/index.json` and flatten the on-disk `hot[<kind>][]`
@@ -81,37 +78,3 @@ export function searchMemory(input) {
         };
     });
 }
-/**
- * Run the search + optionally compress the joined match text via headroom.
- * The structured `matches` array is the canonical output. The
- * `compressedResults` is a derived view for LLM-side prompt assembly.
- */
-export async function searchMemoryWithResults(input, options = {}) {
-    const matches = searchMemory(input);
-    if (options.compressResults !== true) {
-        return { matches, compressedResults: null };
-    }
-    const projectRoot = input.projectRoot ?? process.cwd();
-    const prefs = loadPreferences(projectRoot).headroom;
-    const joinedText = matches.map((m) => `${m.name}: ${m.description}`).join('\n');
-    const joinedBytes = Buffer.byteLength(joinedText, 'utf8');
-    const decision = shouldCompressResults(prefs, joinedBytes, 'memorySearch');
-    if (!decision.compress) {
-        return { matches, compressedResults: null };
-    }
-    const result = await compressPrompt(joinedText, decision.mode);
-    if (result.warning !== null || result.compressedPrompt === null) {
-        return { matches, compressedResults: null };
-    }
-    return {
-        matches,
-        compressedResults: {
-            mode: decision.mode,
-            originalSize: result.originalSize,
-            compressedSize: result.compressedSize,
-            compressionRatio: result.compressionRatio,
-            compressedText: result.compressedPrompt,
-            warning: result.warning
-        }
-    };
-}

package/dist/services/preferences/preferences-service.d.ts

@@ -16,14 +16,11 @@ export declare function writeAtomic(filePath: string, prefs: ProjectPreferences)
  * current shape. Returns the migrated object + a list of changes
  * applied (for surfacing in the CLI envelope and the migration log).
  *
- * The v1 → v2 mapping fills in the `headroom.perTouchpoint` sub-keys
- * that v1 didn't track: each touchpoint gets the v1 `headroom.defaultMode`
- * (or 'balanced' if absent), and `compressMinBytes` is preserved
- * verbatim. v1's `agentShieldPrompt` and `loopAutonomousEnabled` were
- * already present, so they carry through unchanged. The `fanout` field
- * is new in v2 — it defaults to `{ defaultMode: 'fan-out' }` so the
- * pre-v2 behavior (peak-code SKILL instructed fan-out when ≥ 2 leaves)
- * is preserved.
+ * v1's `agentShieldPrompt` and `loopAutonomousEnabled` were already
+ * present, so they carry through unchanged. The `fanout` field is new
+ * in v2 — it defaults to `{ defaultMode: 'fan-out' }` so the pre-v2
+ * behavior (peak-code SKILL instructed fan-out when ≥ 2 leaves) is
+ * preserved.
  */
 export type MigrateResult = {
     readonly fromVersion: string;

package/dist/services/preferences/preferences-service.js

@@ -135,25 +135,6 @@ export function migratePreferences(projectRoot, opts = {}) {
         };
     }
     const changes = [];
-    // v1 → v2: headroom.perTouchpoint fill-in (v1 only had defaultMode).
-    const headroomRaw = (raw.headroom ?? {});
-    const defaultMode = isHeadroomModeString(headroomRaw.defaultMode)
-        ? headroomRaw.defaultMode
-        : 'balanced';
-    const perTouchpoint = {
-        subAgentDispatch: defaultMode,
-        memorySearch: defaultMode,
-        retrospectiveSearch: defaultMode,
-        doctorScan: defaultMode,
-        doctorRoute: 'conservative'
-    };
-    changes.push(`headroom.perTouchpoint filled in with defaultMode='${defaultMode}' (v1 lacked per-touchpoint overrides)`);
-    if (typeof headroomRaw.compressMinBytes === 'number') {
-        changes.push(`headroom.compressMinBytes preserved at ${headroomRaw.compressMinBytes}`);
-    }
-    else {
-        changes.push(`headroom.compressMinBytes defaulted to 4096`);
-    }
     // v1 → v2: fanout block was absent; v2 needs it. Default to the
     // pre-v2 behavior (fan-out when ≥ 2 leaves).
     if (raw.fanout === undefined) {
@@ -176,14 +157,6 @@ export function migratePreferences(projectRoot, opts = {}) {
     // fully-valid v2 ProjectPreferences.
     const migrated = mergePreferences(DEFAULT_PREFERENCES, {
         ...raw,
-        headroom: {
-            enabled: typeof headroomRaw.enabled === 'boolean' ? headroomRaw.enabled : true,
-            defaultMode,
-            perTouchpoint,
-            compressMinBytes: typeof headroomRaw.compressMinBytes === 'number'
-                ? headroomRaw.compressMinBytes
-                : 4096
-        },
         fanout: typeof raw.fanout === 'object' && raw.fanout !== null && !Array.isArray(raw.fanout)
             ? (raw.fanout.defaultMode === 'serial'
                 ? { defaultMode: 'fan-out' }
@@ -203,6 +176,3 @@ export function migratePreferences(projectRoot, opts = {}) {
         written
     };
 }
-function isHeadroomModeString(value) {
-    return value === 'balanced' || value === 'aggressive' || value === 'conservative';
-}

package/dist/services/preferences/preferences-types.d.ts

@@ -22,27 +22,6 @@ export type UaPromptDecision = 'unset' | 'skip-this-session' | 'skip-forever';
  *   - 'lax'      — always downgrade to previous level (faster, riskier)
  */
 export type ClassifyConservatism = 'default' | 'strict' | 'lax';
-/**
- * Per-touchpoint headroom-AI mode override.
- * Spec §7.4 — default 'balanced'.
- */
-export type HeadroomMode = 'balanced' | 'aggressive' | 'conservative';
-export interface HeadroomPreferences {
-    /** Whether headroom integration is enabled globally. Default: true */
-    readonly enabled: boolean;
-    /** Default mode if a touchpoint doesn't override. Default: 'balanced' */
-    readonly defaultMode: HeadroomMode;
-    /** Per-touchpoint mode overrides */
-    readonly perTouchpoint: {
-        subAgentDispatch: HeadroomMode;
-        memorySearch: HeadroomMode;
-        retrospectiveSearch: HeadroomMode;
-        doctorScan: HeadroomMode;
-        doctorRoute: HeadroomMode;
-    };
-    /** Minimum joined-result byte count before search-touchpoint compression runs. Default: 4096. */
-    readonly compressMinBytes: number;
-}
 export interface ClassifyRuleOverrides {
     /** File count threshold above which a task is promoted to 'feature' */
     readonly feature_threshold_files?: number;
@@ -101,7 +80,6 @@ export interface ProjectPreferences {
     readonly agentShieldPrompt: UaPromptDecision;
     readonly classifyConservatism: ClassifyConservatism;
     readonly classifyRules: ClassifyRuleOverrides;
-    readonly headroom: HeadroomPreferences;
     readonly swarmSpeculative: SwarmSpeculativePreferences;
     /** Loop Autonomous (L4 14.5) toggle. Default: false — never auto-enable. */
     readonly loopAutonomousEnabled: boolean;

package/dist/services/preferences/preferences-types.js

@@ -28,18 +28,6 @@ export const DEFAULT_PREFERENCES = {
         feature_threshold_lines: 100,
         runtime_clean_grace_hours: 24,
     },
-    headroom: {
-        enabled: true,
-        defaultMode: 'balanced',
-        perTouchpoint: {
-            subAgentDispatch: 'balanced',
-            memorySearch: 'balanced',
-            retrospectiveSearch: 'balanced',
-            doctorScan: 'balanced',
-            doctorRoute: 'conservative',
-        },
-        compressMinBytes: 4096,
-    },
     swarmSpeculative: {
         enabled: true,
         maxConcurrent: 3,

package/dist/services/retrospective/retrospective-search-service.d.ts

@@ -35,21 +35,3 @@ export interface RetrospectiveSearchResult {
  * runs (cheaper to filter, then fuzzy on a smaller set).
  */
 export declare function searchRetrospective(input: RetrospectiveSearchInput): RetrospectiveSearchResult[];
-export interface CompressedRetrospectiveResultsEnvelope {
-    readonly mode: 'balanced' | 'aggressive' | 'conservative';
-    readonly originalSize: number;
-    readonly compressedSize: number;
-    readonly compressionRatio: number;
-    readonly compressedText: string;
-    readonly warning: string | null;
-}
-export interface RetrospectiveSearchWithResults {
-    readonly matches: RetrospectiveSearchResult[];
-    readonly compressedResults: CompressedRetrospectiveResultsEnvelope | null;
-}
-export interface RetrospectiveSearchOptions {
-    /** When true, compress joined match text via headroom-ai. Falls back
-     *  silently to the structured `matches` array on failure. */
-    readonly compressResults?: boolean;
-}
-export declare function searchRetrospectiveWithResults(input: RetrospectiveSearchInput, options?: RetrospectiveSearchOptions): Promise<RetrospectiveSearchWithResults>;

package/dist/services/retrospective/retrospective-search-service.js

@@ -73,35 +73,3 @@ export function searchRetrospective(input) {
         };
     });
 }
-import { compressPrompt as compressPromptForRetro } from '../context/headroom-client.js';
-import { loadPreferences as loadPreferencesForRetro } from '../preferences/preferences-service.js';
-import { shouldCompressResults as shouldCompressResultsForRetro } from '../context/headroom-prefs.js';
-export async function searchRetrospectiveWithResults(input, options = {}) {
-    const matches = searchRetrospective(input);
-    if (options.compressResults !== true) {
-        return { matches, compressedResults: null };
-    }
-    const projectRoot = input.projectRoot ?? process.cwd();
-    const prefs = loadPreferencesForRetro(projectRoot).headroom;
-    const joinedText = matches.map((m) => `${m.title}: ${m.summary}`).join('\n');
-    const joinedBytes = Buffer.byteLength(joinedText, 'utf8');
-    const decision = shouldCompressResultsForRetro(prefs, joinedBytes, 'retrospectiveSearch');
-    if (!decision.compress) {
-        return { matches, compressedResults: null };
-    }
-    const result = await compressPromptForRetro(joinedText, decision.mode);
-    if (result.warning !== null || result.compressedPrompt === null) {
-        return { matches, compressedResults: null };
-    }
-    return {
-        matches,
-        compressedResults: {
-            mode: decision.mode,
-            originalSize: result.originalSize,
-            compressedSize: result.compressedSize,
-            compressionRatio: result.compressionRatio,
-            compressedText: result.compressedPrompt,
-            warning: result.warning
-        }
-    };
-}

package/dist/services/slice/slice-benchmark-service.d.ts

@@ -1,5 +1,5 @@
 /**
- * Slice benchmark wrapper (slice 2026-06-14-fzf-headroom-rollout).
+ * Slice benchmark wrapper (slice 2026-06-14-fzf-rollout).
  *
  * Wraps `decomposeSlices` with a thin observability layer that records
  * total wall time, codegraph call count, confidence distribution, and

package/dist/services/slice/slice-benchmark-service.js

@@ -1,5 +1,5 @@
 /**
- * Slice benchmark wrapper (slice 2026-06-14-fzf-headroom-rollout).
+ * Slice benchmark wrapper (slice 2026-06-14-fzf-rollout).
  *
  * Wraps `decomposeSlices` with a thin observability layer that records
  * total wall time, codegraph call count, confidence distribution, and

package/dist/services/slice/slice-pick-service.d.ts

@@ -1,7 +1,7 @@
 /**
  * Slice Pick Service — thin wrapper around the generic fzf picker.
  *
- * v2 (slice 2026-06-14-fzf-headroom-rollout): the actual fzf
+ * v2 (slice 2026-06-14-fzf-rollout): the actual fzf
  * integration is now in `src/services/fuzzy-matching/fzf-pick-service.ts`.
  * This file only knows how to format / parse SliceCandidate rows and
  * pick the output path. The algorithm itself (slice-decompose-service)

package/dist/services/slice/slice-pick-service.js

@@ -1,7 +1,7 @@
 /**
  * Slice Pick Service — thin wrapper around the generic fzf picker.
  *
- * v2 (slice 2026-06-14-fzf-headroom-rollout): the actual fzf
+ * v2 (slice 2026-06-14-fzf-rollout): the actual fzf
  * integration is now in `src/services/fuzzy-matching/fzf-pick-service.ts`.
  * This file only knows how to format / parse SliceCandidate rows and
  * pick the output path. The algorithm itself (slice-decompose-service)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "peaks-loop",
-  "version": "4.0.25",
+  "version": "4.0.26",
   "description": "Loop Engineering CLI — workflow primitive / loop guards / evaluators / slice orchestration",
   "author": "SquabbyZ",
   "keywords": [
@@ -99,13 +99,12 @@
     "better-sqlite3": "^12.11.1",
     "commander": "^12.1.0",
     "fzf": "^0.5.2",
-    "headroom-ai": "0.22.4",
     "yaml": "^2.9.0",
     "zod": "^4.4.3",
-    "peaks-loop-internal-runtime": "0.0.10",
-    "peaks-loop-mut": "0.1.23",
-    "peaks-loop-shared-channel": "0.0.27",
-    "peaks-loop-shared": "0.0.59"
+    "peaks-loop-internal-runtime": "0.0.11",
+    "peaks-loop-shared": "0.0.60",
+    "peaks-loop-shared-channel": "0.0.28",
+    "peaks-loop-mut": "0.1.24"
   },
   "devDependencies": {
     "@changesets/cli": "2.31.1",

package/skills/bee/peaks-qa/references/qa-context-governance.md

@@ -21,4 +21,4 @@ PROTOCOL (mandatory):
 
 ## G9 — QA prompt size self-check
 
-Same as RD: 50% soft warn, 75% `CONTEXT_NEAR_LIMIT`, 80% hard reject unless `--force`. QA test plans can grow large; prefer `--use-headroom balanced` for plans > 75%.
+Same as RD: 50% soft warn, 75% `CONTEXT_NEAR_LIMIT`, 80% hard reject unless `--force`. QA test plans can grow large; trim or split into multiple dispatches for plans > 75%.

package/skills/bee/peaks-rd/SKILL.md

@@ -235,9 +235,9 @@ The `karpathy-reviewer` sub-agent now reports its own runtime cost in the JSON e
 
 The main RD loop MUST call `peaks job karpathy-cost-check` after every `peaks request transition --state qa-handoff` and before the next-slice work begins. If the decision kind is `downgraded`, the LLM MUST honor the downgraded `warn` and proceed to the next slice; the `'block'` was an artifact of the reviewer's own cost, not the slice's quality. If the decision kind is `reported` (costRatio > 50, gate not `'block'`), the LLM MAY continue; the sediment will be appended by `peaks memory extract` at handoff time. The full design lives in `.peaks/memory/2026-07-30-karpathy-evaluation-cost-self-review-design.md` (sediment locked 2026-07-30).
 
-## Sub-agent context governance (G7 + G7.7 + G8 + G9 — slice #010)
+## Sub-agent context governance (G7 + G8 + G9 — slice #010)
 
-RD sub-agent prompt template MUST include the G7 path convention + G8.6 share protocol. Detailed protocol: `skills/peaks-code/references/context-governance.md` + `skills/peaks-code/references/headroom-integration.md`.
+RD sub-agent prompt template MUST include the G7 path convention + G8.6 share protocol. Detailed protocol: `skills/peaks-code/references/context-governance.md`.
 
 → see `references/rd-context-governance.md` for the full G7 / G8.6 / G9 protocol + RD sub-agent prompt template.
 

package/skills/bee/peaks-rd/references/rd-context-governance.md

@@ -31,6 +31,6 @@ PROTOCOL (mandatory):
 
 Before dispatching a sub-agent, RD self-checks prompt size:
 - < 50%: pass through.
-- 50-75%: soft warn (consider `--use-headroom`).
-- 75-80%: soft warn + `warnings: ["CONTEXT_NEAR_LIMIT"]` (mandatory suggest `--use-headroom`).
+- 50-75%: soft warn (consider trimming the prompt).
+- 75-80%: soft warn + `warnings: ["CONTEXT_NEAR_LIMIT"]` (trim or split into multiple dispatches).
 - 80%+: reject (CLI 兜底 returns `code: "PROMPT_TOO_LARGE"`). Use `--force` at CLI only when overriding; hook layer will still reject (RL-30).

package/skills/bee/peaks-txt/SKILL.md

@@ -305,7 +305,7 @@ The TXT handoff summarizes the slice by listing the artifact metas + the share e
 
 ### G9 — TXT prompt size self-check
 
-Same G9 threshold table. TXT handoff messages can grow large when the slice has many batches; use `--use-headroom` proactively for handoff prompts > 50%.
+Same G9 threshold table. TXT handoff messages can grow large when the slice has many batches; trim or split the handoff prompt proactively for prompts > 50%.
 
 
 ## L2 surface reference (post rid-l2-extended)

package/skills/bee/peaks-ui/SKILL.md

@@ -361,7 +361,7 @@ PROTOCOL (mandatory):
 
 ### G9 — UI prompt size self-check
 
-Same as RD/QA. Use `--use-headroom` proactively.
+Same as RD/QA. Trim or split the prompt proactively.
 
 
 ## L2 surface reference (post rid-l2-extended)

package/skills/peaks-code/SKILL.md

@@ -302,7 +302,7 @@ After final validation, refresh project-local standards via `peaks standards ini
 
 `peaks codegraph affected` is an optional project-analysis enhancement (untrusted supporting evidence) for role handoff. Code must not treat codegraph output as approval for scope, design, or QA verdict. Never mutate agent settings / hooks from codegraph; do not commit `.codegraph/` artifacts. RD writes `.peaks/_runtime/<sessionId>/rd/codegraph-context.md`; QA / TXT consume the same envelope. → `references/codegraph-orchestration.md`.
 
-## Sub-agent context governance (G7 + G7.7 + G8 + G9 — slice #010)
+## Sub-agent context governance (G7 + G8 + G9 — slice #010)
 
 Main LLM reducer sees metadata-only view (~200 chars/sub-agent); on-demand `Read` for full content. Threshold table: 50% soft warn, 75% `CONTEXT_NEAR_LIMIT`, 80% hard reject (CLI + hook double-guard). → `references/context-governance.md`.
 

package/skills/peaks-code/references/context-governance.md

@@ -1,7 +1,7 @@
-# Context Governance — G7 + G7.7 + G8 + G9 protocol details
+# Context Governance — G7 + G8 + G9 protocol details
 
-> Slice #010 (G7 + G7.7 + G8 + G9 context-governance push).
-> See: `.peaks/memory/sub-agent-context-minimal-occupation.md` + `sub-agent-shared-channel-cross-completion.md` + `sub-agent-headroom-forced-compression-gate.md` for the red lines.
+> Slice #010 (G7 + G8 + G9 context-governance push).
+> See: `.peaks/memory/sub-agent-context-minimal-occupation.md` + `sub-agent-shared-channel-cross-completion.md` for the red lines.
 
 ## G7 — sub-agent context minimal-occupation (metadata-only + 按需 Read)
 
@@ -56,26 +56,6 @@ On completion:
 
 3000-5000× improvement. Main LLM full-slice context net increase: < 10KB for 5 batches × 6 sub-agents.
 
-## G7.7 — headroom-ai integration (opt-in)
-
-### `--use-headroom` flag
-
-Opt-in flag on `peaks sub-agent dispatch`. Default `false` (G7 metadata-only remains the default).
-
-### Mode table
-
-| Mode | tokenBudget | Use case |
-|---|---|---|
-| `balanced` (default) | promptSize * 0.40 / 4 | General sub-agent dispatch |
-| `aggressive` | promptSize * 0.20 / 4 | Last-resort large prompt |
-| `conservative` | promptSize * 0.70 / 4 | Sensitive code analysis |
-
-### Failure mode (RL-22d / RL-32)
-
-- headroom daemon dead / proxy unreachable / times out
-- → `code: "HEADROOM_UNAVAILABLE"` warning + G7 metadata-only fallback
-- → NOT blocking (warn, then continue dispatch)
-
 ## G8 — cross sub-agent shared channel
 
 ### Path convention
@@ -132,8 +112,8 @@ PROTOCOL (mandatory):
 
 | Threshold | Prompt size | Behavior |
 |---|---|---|
-| 50% (early warn) | ≥ 128KB | Soft warning, suggest `--use-headroom` |
-| **75% (user red line)** | ≥ 192KB | Soft warn + mandatory suggest `--use-headroom`; `warnings: ["CONTEXT_NEAR_LIMIT"]` |
+| 50% (early warn) | ≥ 128KB | Soft warning, suggest trimming the prompt |
+| **75% (user red line)** | ≥ 192KB | Soft warn + mandatory trim/split suggestion; `warnings: ["CONTEXT_NEAR_LIMIT"]` |
 | **80% (hard reject)** | ≥ 204KB | Hard reject `code: "PROMPT_TOO_LARGE"`; `--force` allowed at CLI |
 | **90% (emergency)** | ≥ 230KB | Hard reject `code: "PROMPT_EMERGENCY"`; `--force` STILL rejects at 90% (no override) |
 
@@ -149,7 +129,7 @@ PROTOCOL (mandatory):
 
 ## AC mapping
 
-- AC-38..AC-43 (G7) + AC-44..AC-46 (G7.7) + AC-47..AC-49 (G8) + AC-50..AC-65 (G9)
+- AC-38..AC-43 (G7) + AC-47..AC-49 (G8) + AC-50..AC-65 (G9)
 - See PRD §Acceptance criteria.
 
 ---
@@ -171,13 +151,6 @@ Main LLM view format (G7.4.e):
 - qa-perf → .../artifacts/003-qa-perf-001.md (5KB, sha256:ghi789) summary: "p95 latency target ≤ 200ms"
 ```
 
-### G7.7 — headroom-ai integration (opt-in compression)
-
-> Body of `### G7.7`. If a sub-agent prompt is too large even after G7 metadata-only (e.g. 1MB artifact description, 5MB mid-prompt analysis), use `--use-headroom`:
-- Default `false` (G7 remains default).
-- Modes: `balanced` (default) | `aggressive` | `conservative`.
-- Failure: `HEADROOM_UNAVAILABLE` warning + G7 metadata-only fallback (NOT blocking).
-
 ### G8 — cross sub-agent shared channel (dispatcher-mediated indirect signal)
 
 > Body of `### G8`. Sub-agent A's completion **immediately** writes a shared entry; sub-agent B (still in flight) can read shared entries from sibling sub-agents. **This is NOT peer-to-peer messaging.** The dispatcher stores, the sub-agents read/write; A and B never directly talk.
@@ -192,7 +165,7 @@ Main LLM view format (G7.4.e):
 
 | Threshold | Prompt size | Behavior |
 |---|---|---|
-| 50% (early warn) | ≥ 128KB | Soft warning, suggest `--use-headroom` |
+| 50% (early warn) | ≥ 128KB | Soft warning, suggest trimming the prompt |
 | **75% (user red line)** | ≥ 192KB | Soft warn + `warnings: ["CONTEXT_NEAR_LIMIT"]` |
 | **80% (hard reject)** | ≥ 204KB | Hard reject `code: "PROMPT_TOO_LARGE"`; `--force` allowed at CLI |
 | 90% (emergency) | ≥ 230KB | Hard reject + `contextWarning: 'high'` |

package/skills/peaks-code/references/envelope-contract.md

@@ -33,19 +33,14 @@ Top-level fields (post-2.1.0):
 - `envelopeVersion: '2.1.0'`
 - `role: string` — sub-agent role string
 - `ide: string` — detected IDE label (`claude-code` etc.)
-- `originalPromptSize: number` — bytes of the un-compressed prompt
-- `promptSize: number` — bytes after headroom compression (or
-  equal to `originalPromptSize` when `--use-headroom` is off or
-  unavailable)
+- `originalPromptSize: number` — bytes of the caller-supplied prompt
+- `promptSize: number` — bytes of the composed dispatch prompt
 - `toolCall: { name: string; args: Record<string, unknown> }` —
   the per-IDE tool-call descriptor the LLM must execute
 - `dispatchRecordPath: string` — absolute path to the dispatch
   record on disk
 - `batchId: string` — uuid-like opaque token grouping one batch
 - `dispatchedInBatch: number` — current count after this dispatch
-- `headroomCompressed: boolean` — true if headroom-ai actually
-  reduced the prompt
-- `headroomResult: { mode, compressed, compressionRatio, tokensSaved, warning } | null`
 - `forcedAt: string | null` — ISO8601 when `--force` overrode
   the G9 hard-reject tier
 - `contextImpact: { promptBytes, artifactBytes, totalBytes }`

package/skills/peaks-code/references/sub-agent-dispatch.md

@@ -51,8 +51,8 @@ peaks sub-agent dispatch <role> --prompt <text> [--request-id <rid>] [--session-
 > scrollback. The dispatch record on disk (gitignored under
 > `.peaks/_sub_agents/`) keeps the prompt for the sub-agent to read;
 > CLI stdout stays metadata-only. Surface `originalPromptSize` +
-> `promptSize` so the LLM-side runner can still reason about headroom
-> without seeing the content. The prompt content the LLM should pass
+> `promptSize` so the LLM-side runner can still reason about the size
+> delta without seeing the content. The prompt content the LLM should pass
 > through lives in `data.toolCall.args.prompt` (the IDE-arg the
 > consumer passes to the tool), NOT in the outer envelope.
 
@@ -77,8 +77,6 @@ peaks sub-agent dispatch <role> --prompt <text> [--request-id <rid>] [--session-
     "dispatchRecordPath": ".peaks/_sub_agents/2026-06-06-session-5b1095/dispatch-002-2026-06-07-...-...json",
     "batchId": "<uuid>",
     "dispatchedInBatch": 3,
-    "headroomCompressed": false,
-    "headroomResult": null,
     "forcedAt": null,
     "contextImpact": {
       "promptBytes": 4321,

package/dist/services/context/headroom-client.d.ts

@@ -1,34 +0,0 @@
-export type HeadroomMode = 'balanced' | 'aggressive' | 'conservative';
-export interface HeadroomResult {
-    readonly compressed: boolean;
-    readonly originalSize: number;
-    readonly compressedSize: number;
-    readonly compressionRatio: number;
-    readonly mode: HeadroomMode;
-    /** `'HEADROOM_UNAVAILABLE'` on fallback; `null` on success. */
-    readonly warning: string | null;
-    /** Compressed prompt body. `null` if no compression happened. */
-    readonly compressedPrompt: string | null;
-    /** Tokens saved (from the SDK). 0 on fallback. */
-    readonly tokensSaved: number;
-}
-/**
- * Compress a prompt via headroom-ai. The `fallback: true` option is
- * non-negotiable: if the proxy daemon is unavailable, the SDK returns
- * `result.compressed = false` and the original messages; we surface
- * that as `HEADROOM_UNAVAILABLE` warning + G7 metadata-only fallback.
- */
-export declare function compressPrompt(prompt: string, mode?: HeadroomMode): Promise<HeadroomResult>;
-/**
- * Bridge interface: when `--use-headroom` is set, share entries written
- * via `peaks sub-agent share` MAY also flow through headroom's
- * `SharedContext`. Slice #010 implements a peak-internal shared channel
- * (see `shared-channel.ts`); the headroom-side `SharedContext` is a
- * separate concept that future slices can layer on. For now this
- * function is a stub that returns the peak-internal channel ID, which
- * is enough to demonstrate the bridge contract.
- */
-export declare function buildSharedContextBridge(batchId: string): {
-    peakChannelId: string;
-    headroomContextId: string;
-};