[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C522FSMS4XKSPB1DZ3D6FM":3,"skill-same:01M7C522FSMS4XKSPB1DZ3D6FM":57},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M7C522FSMS4XKSPB1DZ3D6FM","session-handoff",null,"Write handoff, spawn successor, answer grill, verify orientation. Triggers: hand off, phase change. NOT on context%.","EtanHey\u002Fgolems","EtanHey","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F93338119?v=4","01M7C4SKC9HJC3B9TAX0D9MZ1E","Etan Heyman","01M7C4SKCT59Z49TTN4WWDDEYD",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fsession-handoff\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: session-handoff\ndescription: \"Write handoff, spawn successor, answer grill, verify orientation. Triggers: hand off, phase change. NOT on context%.\"\n---\n\n# Session Handoff — Structured Agent Continuation\n\nA handoff fails when the outgoing agent sends an unstructured wall of text, omits monitoring, or waits for the user to answer the successor's orientation questions. The outgoing agent already owns that context and must transfer it directly.\n\n---\n\n## THE HANDOFF PROTOCOL (5 steps, in order)\n\n### Step 1: Write the Handoff File\n\n**NEVER send handoff as raw cmux text.** Write a structured file that the new agent reads.\n\nLocation: `docs.local\u002Fhandoffs\u002Fhandoff-{date}-{session-summary}.md`\n\nTemplate:\n\n```markdown\n# Session Handoff — {date}\n\n## Outgoing Agent\n- Agent: {name} (e.g., orcClaude)\n- Context: {token_count}\u002F{model_max} ({utilization}%)\n- Session ID: {id}\n- Duration: {time}\n\n## Session Intent\n{Exact user quotes about what they want to achieve — verbatim, not paraphrased}\n\n## Decisions Made\n| Decision | Why | Who Decided |\n|----------|-----|-------------|\n| {decision} | {rationale} | {user\u002Fagent} |\n\n## User Corrections (from \u002Ffrustration-capture)\n| Correction | Category | Importance | Quote |\n|-----------|----------|------------|-------|\n| {what was wrong → what's right} | {category} | {7-10} | \"{exact words}\" |\n\n## Full-Relay (MANDATORY when a weave\u002Fretro exists for this window)\n- Weave doc: {path — successor MUST read it IN FULL, not just this summary}\n- Corrections sweep: brain_search tag sweep ({tags}) — successor reads ALL, ACKs item-by-item\n- §EDITS application table state: {applied\u002Fdispatched per item, links}\n\n## Agent State\n| Agent | T3 thread id \u002F taskId (cmux: agent_id) | Surface (cmux only) | Task | Status | Context % | Launcher |\n|-------|----------|---------|------|--------|-----------|----------|\n| {name} | {thread id \u002F taskId, or agent_id} | surface:{N} or n\u002Fa | {task} | {idle\u002Fready\u002Flearning\u002Fworking\u002Fblocked\u002Fdone\u002Fsigned-off} | {%} | {launcher -flags} |\n\n## Open PRs\n| Repo | PR # | Status | What |\n|------|------|--------|------|\n| {repo} | #{N} | {open\u002Freviewing\u002Fmerged} | {1-line} |\n\n## Active Collabs\n| File | Status | Current Ask |\n|------|--------|-------------|\n| {path} | {active\u002Fblocked} | {what's needed} |\n\n## Current State\n{What's working, what's broken, what's in progress — be specific}\n\n## Next Steps (ordered by priority)\n1. {highest priority task}\n2. {second priority}\n3. ...\n\n## Live-Watch Transfer List (MANDATORY — incomplete brief without this)\n\nEnumerate **every** cron, monitor, harvest loop, and in-flight watch that **dies with the outgoing session** and must be recreated or explicitly retired:\n\n| Watch | ID \u002F schedule | Owner after handoff | Recreate? |\n|-------|---------------|---------------------|-----------|\n| 30-min stall-check cron | e.g. `06827a10` 23,53 * * * * | successor orc | YES |\n| collab-milestones fswatch | task id | successor | YES |\n| {harvest loop} | {id} | {named owner} | YES\u002FNO |\n\nIf it isn't listed here, it doesn't transfer — and the successor won't know it existed.\n\n## Succession broadcast (prevent duplicate handoffs)\n\nAt generation handover:\n\n1. **Mark yourself outgoing.** T3: skip this step (t3-native.md documents no thread-title step); the SUCCESSION line below is the signal. Terminal (cmux) seats only: rename your pane to `{name}-OUTGOING` or `{name}-RETIRED` via `update_surface({action:\"rename\", surface, title})`.\n2. **Broadcast one line** to the active collab: `SUCCESSION: {outgoing} → {successor} | handoff: {absolute path} | watches: {N} items transferred`.\n3. Successor ACKs the handoff path + watch table before spawning parallel work; this prevents two orcs from creating duplicate generation handoffs.\n\n## What the New Agent Should Do First\n1. brain_search(\"handoff session {date}\") to verify this was stored\n2. Read this file\n3. Check active agents. T3: the open `delegate_task` taskIds and thread ids in this file. cmux seats: `list_agents`, or `list_agents({mine:true})` for your own children\n4. Resume monitoring. T3: completion notifications wake you, no polling. cmux seats: `wait_for`, or `list_agents({agent_ids:[id], detail:\"full\"})` for a state snapshot\n5. Pick up from Next Steps #1\n```\n\n### Step 2: brain_store the Handoff\n\n```\nbrain_store(\n  content: \"SESSION HANDOFF {date}: {1-paragraph summary}. Handoff file: {path}. Key state: {active agents}, {open PRs}, {next priority}. User corrections captured: {count}.\",\n  tags: [\"handoff\", \"session-end\", \"\u003Cproject>\"],\n  importance: 8\n)\n```\n\nThis ensures the new agent can find the handoff via brain_search even if the file path changes.\n\n### Step 3: Spawn the New Agent\n\n**T3 (default):** a successor lead is a new thread, `t3_thread_launch` with \"Read and follow \u003Chandoff path>\"; a continuation worker is an async `delegate_task` with the same one-line prompt. Record the successor's thread id (and `taskId` for a delegate) in the handoff notes and the BrainLayer entry. `t3_thread_launch` has no idempotency key: after a lost response, check `t3_thread_list` before retrying.\n\n**cmux (terminal seats only):**\n\n```\n1. `spawn_agent({repo, cli, model, prompt})`\n2. Capture the returned `agent_id`\n3. `wait_for({agent_id, target_state:\"ready\", timeout_ms:120000})`\n4. If `wait_for` times out but `read_screen` shows a prompt, treat it as FR-06 parser ambiguity and continue\n```\n\n**CRITICAL:** Spawn the continuation agent through `spawn_agent`, not by typing a raw launcher into a fresh pane. Keep the `agent_id` in the handoff notes so the next monitor loop does not depend on a remembered surface number.\n\n### Step 4: Answer the Grill (OUTGOING agent's job)\n\nThe new agent will ask orientation questions (\"What's the current state?\", \"What should I focus on?\"). The OUTGOING agent must answer these — NOT the user.\n\n**T3:** a delegate child's question surfaces as a pending request: read it with `t3_pending_request_read` and answer with `t3_pending_request_respond`. A launched successor thread is answered with `t3_thread_send`, only to a thread id from your brief or one returned by your own `t3_thread_launch`\u002F`delegate_task` call in this handoff. Never invent or guess one. Track it by that id.\n\n**cmux process (terminal seats only):**\n1. Track the new agent by `agent_id`, not by surface index\n2. Use `wait_for({agent_id, target_state:\"idle\", timeout_ms:120000})` or `list_agents({agent_ids:[agent_id], detail:\"full\"})` to catch the moment it is waiting for input\n3. When the new agent asks questions → answer via `send_to({agent_id, text, press_enter:true})`\n4. Continue answering until the new agent says \"oriented\" or starts working independently\n5. ONLY THEN tell the user the handoff is complete\n\nThe outgoing agent must answer from its existing context instead of delegating orientation back to the user.\n\n### Step 5: Verify and Release\n\nBefore declaring handoff complete:\n\n1. **New agent is working** — T3: one `t3_thread_read` of the successor thread shows it working. cmux: `wait_for` or `list_agents({agent_ids:[id], detail:\"full\"})` shows active progress, not a waiting prompt\n2. **Monitoring is active** — the new agent is tracked by its T3 thread id (cmux: `agent_id`) and has a state-check path (T3: completion notification or one `t3_thread_read`; cmux: `wait_for`)\n3. **Corrections are stored** — brain_search(\"user-correction\") returns the corrections from this session\n4. **Handoff file exists** — Read the file to verify it was written correctly\n\nThen: `brain_store(\"Handoff complete to {new-agent} ({thread id or agent_id}). Verified: working, monitoring active, corrections stored.\")` and notify user.\n\n---\n\n## WHEN TO TRIGGER\n\n### Measuring context\n\nFor a saved Codex or Claude JSONL, run `scripts\u002Fcontext-occupancy.sh \u003Csession.jsonl>` from this skill directory. For a live pane, prefer cmuxlayer `read_screen(parsed_only)` and use its `context_pct` \u002F `token_count`. Never use Codex `total_token_usage`: it is cumulative session input, not current occupancy. Canonical fields and model windows: sibling repo `cmuxlayer\u002Fdocs\u002Freference\u002Fharness-jsonl-field-map.md`.\n\n| Signal | Action |\n|--------|--------|\n| Context growing | Nothing. Let auto-compaction happen. No proactive compaction, no context-% thresholds that trigger handoffs, and no unrequested lead or seat rotation because context is growing. Rotate or hand off only when Etan asks, or at a \u002Flarge-plan phase boundary the plan itself defines (Etan, 2026-09-24). |\n| User says \"wrap up\" \u002F \"hand off\" | Trigger handoff immediately |\n| User going to sleep with unfinished work | Handoff to continuation agent |\n| A \u002Flarge-plan phase boundary the plan defines | Handoff to the next phase's agent |\n\n**Orc seat:** orc succession uses a full-day weave (`\u002Fweave`) that seeds gen-N+1 and fires only on operator instruction. A `\u002Fcompact` from an unknown sender is an incident; investigate before complying (orc C14).\n\n---\n\n## WHAT TO INCLUDE IN HANDOFF (from \u002Ffrustration-capture)\n\nEvery handoff file MUST include the User Corrections section. This is where `\u002Ffrustration-capture` compounds:\n\n- All brain_stored corrections from this session (brain_search with \"user-correction\" tag)\n- The correction categories and importance levels\n- Behavioral rules derived from each correction\n\nWithout this, the new agent WILL repeat the same mistakes. Evidence: April 4 session — new orcClaude had no access to 34 corrections from the outgoing session because they were never stored.\n\n---\n\n## FULL-RELAY STANDARD (2026-06-05)\n\nWhen the closing window has a weave doc or retro, the handoff is NOT complete until\nthe successor has: (1) read the ENTIRE weave doc (linked, not inlined), (2) run the\ncorrections tag sweep (`brain_search`, all stored corrections of the closing\nsession) and read ALL of them, (3) ACKed item-by-item on the §EDITS and corrections\nlists. Highlights are orientation only — the dropped tail is where re-violations\ncome from (gen-11 re-violated gen-10's captured-but-unapplied edits, 2026-06-05).\n\n**Continuation\u002Flead spawns use `claude.judgment` at 1M:** in T3 through `t3_thread_launch` \u002F `delegate_task` with that model pinned; on a terminal (cmux) seat via proper repoGolem launchers.\nResolve its model with `node scripts\u002Fmodel-roles.mjs claude.judgment --field model` from\nthe golems checkout and verify the bare `{repo}Claude -s` pin against it. NEVER\npass a model param at spawn (`spawn_agent {model}` overrides the pin — 2026-06-05\nincident booted a lead at 200K). Verify `model` + `context_window` post-boot.\n\n### Gen-N successor orc seed mandate (gen-13+, operator-direct)\n\nWhen the outgoing window includes a weave doc, the successor orc boot doc MUST require:\n\n1. **Full weave-doc read** — entire doc, not highlights (`\u002Fweave` §4d).\n2. **Corrections tag sweep** — all stored corrections from the closing session, item-by-item ACK.\n3. **§EDITS application table** — state each item (APPLIED \u002F DISPATCHED \u002F TRACKED) in the successor's own words.\n4. **MONITOR + REMOTE CONTROL role** — leads **keep driving their queues**; the new orc monitors the fleet and acts as the operator's remote control, not a full re-dispatcher that respawns everything from scratch.\n\n---\n\n## ANTI-PATTERNS\n\n| Don't | Do Instead |\n|-------|-----------|\n| Send handoff as raw cmux text | Write structured file, new agent reads it |\n| Wait for user to answer grill | Outgoing agent answers from its own context (Step 4) |\n| Forget to set up monitoring | BEFORE telling user handoff is done: T3, record the successor's thread id \u002F `taskId` and confirm it is working with one `t3_thread_read`; terminal (cmux) seats only, establish `wait_for` \u002F `list_agents(detail:\"full\")` coverage |\n| Skip brain_store | Always store — file can be deleted, BrainLayer persists |\n| Start new session fresh without handoff | Even a 5-line handoff is better than cold start |\n| Declare handoff \"complete\" without verification | Check: new agent working, monitoring active, corrections stored (Step 5) |\n| Compact or hand off because context crossed a % | Let auto-compaction happen; hand off only when Etan asks or at a \u002Flarge-plan phase boundary |\n| Relay only \"the top things\" from a weave\u002Fretro | Full-relay: link the entire weave doc, mandate full read + item-by-item ACK |\n| Spawn continuation with an unpinned or wrong model | T3: `t3_thread_launch` \u002F `delegate_task` with the resolved `claude.judgment` model pinned; terminal (cmux) seats: launcher-only (`{repo}Claude -s`). Verify the model and 1M context after boot |\n| Assume a `-c` resume continues the pre-death turn's duties | Treat resume as fresh — re-derive duties from durable artifacts; encode standing orders as machinery (watch-v6 pattern) |\n\n---\n\n## INTEGRATION\n\n| Skill | How Session Handoff Uses It |\n|-------|---------------------------|\n| `\u002Ffrustration-capture` | Corrections section in handoff file — most valuable knowledge to transfer |\n| `\u002Fagent-routing` | Agent State table includes routing assignments (which agent has which workers) |\n| `\u002Forc` | W6 is the minimal version — this skill is the complete version |\n| `\u002Fcmux-agents` | Spawn-and-verify pattern from `spawn_agent` \u002F `wait_for` for the new agent |\n| `\u002Fnever-fabricate` | Verify handoff file contents before declaring complete |\n\n---\n\n### Inherited-citation suspect rule\n\nAfter consuming a session-handoff doc OR resuming post-compaction, all file:line\ncitations inherited from the prior session are SUSPECT until re-Read.\n\nThe receiving agent MUST:\n1. In Phase 0, list every inherited file:line cite with the tag\n   `[SUSPECT: inherited from {compaction|handoff}]`.\n2. Re-Read each cited path before referencing it in any deliverable. If the file\n   doesn't exist or doesn't contain the claimed content, do NOT silently drop —\n   write a \"FABRICATED IN PRIOR BRIEFING: \u003Ccite>\" entry to the deliverable AND\n   brain_store with tag:compaction-fabrication.\n3. Apply the same rule to chunk IDs (brainbar-*): re-resolve via brain_expand\n   before acting on the inherited claim.\n\nMechanism: post-compaction summaries are pattern-completed by the compacting\nmodel and have a non-zero fabrication rate.\n\n**Reason:** handoff timing estimates must use actual context telemetry:\n*\"my prior briefing had fabricated paths from a compaction summary\"* —\nreceiving agent caught fabricated paths only because the agent happened to\nverify. This rule makes it deterministic instead of relying on diligence.\n\n---\n\n### Resume-as-fresh doctrine (`-c` resumes) *(gen-13 weave E03)*\n\nA `-c` resumed session does NOT reliably execute the pre-death turn's pending\nduties — treat every resume as a FRESH session: re-derive the duty list from\ndurable artifacts (handoff file, collab, BrainLayer), never from the dying\nsession's intentions.\n\nStanding orders that must survive a restart are encoded as MACHINERY — a\nlaunchd job, or a Monitor that itself fires the action AND writes the collab\nannounce (the proven watch-v6 pattern) — not as agent to-do intentions.\n\n**Reason:** a resumed watch seat can freeze on its duties\n(ritual #2: verify\u002Fre-arm\u002Fcompletion-line never executed, ~95 min unmonitored;\nritual #3: completion line 96 min late, self-caught) [line 1538]. The fix is\nproven structurally in watch v6: *\"the monitor itself fires the ritual (nohup\ndetached) and writes the collab announce — removing me as failure point\"*\n[line 1623].\n","# Session Handoff — Structured Agent Continuation\n\nA handoff fails when the outgoing agent sends an unstructured wall of text, omits monitoring, or waits for the user to answer the successor's orientation questions. The outgoing agent already owns that context and must transfer it directly.\n\n---\n\n## THE HANDOFF PROTOCOL (5 steps, in order)\n\n### Step 1: Write the Handoff File\n\n**NEVER send handoff as raw cmux text.** Write a structured file that the new agent reads.\n\nLocation: `docs.local\u002Fhandoffs\u002Fhandoff-{date}-{session-summary}.md`\n\nTemplate:\n\n```markdown\n# Session Handoff — {date}\n\n## Outgoing Agent\n- Agent: {name} (e.g., orcClaude)\n- Context: {token_count}\u002F{model_max} ({utilization}%)\n- Session ID: {id}\n- Duration: {time}\n\n## Session Intent\n{Exact user quotes about what they want to achieve — verbatim, not paraphrased}\n\n## Decisions Made\n| Decision | Why | Who Decided |\n|----------|-----|-------------|\n| {decision} | {rationale} | {user\u002Fagent} |\n\n## User Corrections (from \u002Ffrustration-capture)\n| Correction | Category | Importance | Quote |\n|-----------|----------|------------|-------|\n| {what was wrong → what's right} | {category} | {7-10} | \"{exact words}\" |\n\n## Full-Relay (MANDATORY when a weave\u002Fretro exists for this window)\n- Weave doc: {path — successor MUST read it IN FULL, not just this summary}\n- Corrections sweep: brain_search tag sweep ({tags}) — successor reads ALL, ACKs item-by-item\n- §EDITS application table state: {applied\u002Fdispatched per item, links}\n\n## Agent State\n| Agent | T3 thread id \u002F taskId (cmux: agent_id) | Surface (cmux only) | Task | Status | Context % | Launcher |\n|-------|----------|---------|------|--------|-----------|----------|\n| {name} | {thread id \u002F taskId, or agent_id} | surface:{N} or n\u002Fa | {task} | {idle\u002Fready\u002Flearning\u002Fworking\u002Fblocked\u002Fdone\u002Fsigned-off} | {%} | {launcher -flags} |\n\n## Open PRs\n| Repo | PR # | Status | What |\n|------|------|--------|------|\n| {repo} | #{N} | {open\u002Freviewing\u002Fmerged} | {1-line} |\n\n## Active Collabs\n| File | Status | Current Ask |\n|------|--------|-------------|\n| {path} | {active\u002Fblocked} | {what's needed} |\n\n## Current State\n{What's working, what's broken, what's in progress — be specific}\n\n## Next Steps (ordered by priority)\n1. {highest priority task}\n2. {second priority}\n3. ...\n\n## Live-Watch Transfer List (MANDATORY — incomplete brief without this)\n\nEnumerate **every** cron, monitor, harvest loop, and in-flight watch that **dies with the outgoing session** and must be recreated or explicitly retired:\n\n| Watch | ID \u002F schedule | Owner after handoff | Recreate? |\n|-------|---------------|---------------------|-----------|\n| 30-min stall-check cron | e.g. `06827a10` 23,53 * * * * | successor orc | YES |\n| collab-milestones fswatch | task id | successor | YES |\n| {harvest loop} | {id} | {named owner} | YES\u002FNO |\n\nIf it isn't listed here, it doesn't transfer — and the successor won't know it existed.\n\n## Succession broadcast (prevent duplicate handoffs)\n\nAt generation handover:\n\n1. **Mark yourself outgoing.** T3: skip this step (t3-native.md documents no thread-title step); the SUCCESSION line below is the signal. Terminal (cmux) seats only: rename your pane to `{name}-OUTGOING` or `{name}-RETIRED` via `update_surface({action:\"rename\", surface, title})`.\n2. **Broadcast one line** to the active collab: `SUCCESSION: {outgoing} → {successor} | handoff: {absolute path} | watches: {N} items transferred`.\n3. Successor ACKs the handoff path + watch table before spawning parallel work; this prevents two orcs from creating duplicate generation handoffs.\n\n## What the New Agent Should Do First\n1. brain_search(\"handoff session {date}\") to verify this was stored\n2. Read this file\n3. Check active agents. T3: the open `delegate_task` taskIds and thread ids in this file. cmux seats: `list_agents`, or `list_agents({mine:true})` for your own children\n4. Resume monitoring. T3: completion notifications wake you, no polling. cmux seats: `wait_for`, or `list_agents({agent_ids:[id], detail:\"full\"})` for a state snapshot\n5. Pick up from Next Steps #1\n```\n\n### Step 2: brain_store the Handoff\n\n```\nbrain_store(\n  content: \"SESSION HANDOFF {date}: {1-paragraph summary}. Handoff file: {path}. Key state: {active agents}, {open PRs}, {next priority}. User corrections captured: {count}.\",\n  tags: [\"handoff\", \"session-end\", \"\u003Cproject>\"],\n  importance: 8\n)\n```\n\nThis ensures the new agent can find the handoff via brain_search even if the file path changes.\n\n### Step 3: Spawn the New Agent\n\n**T3 (default):** a successor lead is a new thread, `t3_thread_launch` with \"Read and follow \u003Chandoff path>\"; a continuation worker is an async `delegate_task` with the same one-line prompt. Record the successor's thread id (and `taskId` for a delegate) in the handoff notes and the BrainLayer entry. `t3_thread_launch` has no idempotency key: after a lost response, check `t3_thread_list` before retrying.\n\n**cmux (terminal seats only):**\n\n```\n1. `spawn_agent({repo, cli, model, prompt})`\n2. Capture the returned `agent_id`\n3. `wait_for({agent_id, target_state:\"ready\", timeout_ms:120000})`\n4. If `wait_for` times out but `read_screen` shows a prompt, treat it as FR-06 parser ambiguity and continue\n```\n\n**CRITICAL:** Spawn the continuation agent through `spawn_agent`, not by typing a raw launcher into a fresh pane. Keep the `agent_id` in the handoff notes so the next monitor loop does not depend on a remembered surface number.\n\n### Step 4: Answer the Grill (OUTGOING agent's job)\n\nThe new agent will ask orientation questions (\"What's the current state?\", \"What should I focus on?\"). The OUTGOING agent must answer these — NOT the user.\n\n**T3:** a delegate child's question surfaces as a pending request: read it with `t3_pending_request_read` and answer with `t3_pending_request_respond`. A launched successor thread is answered with `t3_thread_send`, only to a thread id from your brief or one returned by your own `t3_thread_launch`\u002F`delegate_task` call in this handoff. Never invent or guess one. Track it by that id.\n\n**cmux process (terminal seats only):**\n1. Track the new agent by `agent_id`, not by surface index\n2. Use `wait_for({agent_id, target_state:\"idle\", timeout_ms:120000})` or `list_agents({agent_ids:[agent_id], detail:\"full\"})` to catch the moment it is waiting for input\n3. When the new agent asks questions → answer via `send_to({agent_id, text, press_enter:true})`\n4. Continue answering until the new agent says \"oriented\" or starts working independently\n5. ONLY THEN tell the user the handoff is complete\n\nThe outgoing agent must answer from its existing context instead of delegating orientation back to the user.\n\n### Step 5: Verify and Release\n\nBefore declaring handoff complete:\n\n1. **New agent is working** — T3: one `t3_thread_read` of the successor thread shows it working. cmux: `wait_for` or `list_agents({agent_ids:[id], detail:\"full\"})` shows active progress, not a waiting prompt\n2. **Monitoring is active** — the new agent is tracked by its T3 thread id (cmux: `agent_id`) and has a state-check path (T3: completion notification or one `t3_thread_read`; cmux: `wait_for`)\n3. **Corrections are stored** — brain_search(\"user-correction\") returns the corrections from this session\n4. **Handoff file exists** — Read the file to verify it was written correctly\n\nThen: `brain_store(\"Handoff complete to {new-agent} ({thread id or agent_id}). Verified: working, monitoring active, corrections stored.\")` and notify user.\n\n---\n\n## WHEN TO TRIGGER\n\n### Measuring context\n\nFor a saved Codex or Claude JSONL, run `scripts\u002Fcontext-occupancy.sh \u003Csession.jsonl>` from this skill directory. For a live pane, prefer cmuxlayer `read_screen(parsed_only)` and use its `context_pct` \u002F `token_count`. Never use Codex `total_token_usage`: it is cumulative session input, not current occupancy. Canonical fields and model windows: sibling repo `cmuxlayer\u002Fdocs\u002Freference\u002Fharness-jsonl-field-map.md`.\n\n| Signal | Action |\n|--------|--------|\n| Context growing | Nothing. Let auto-compaction happen. No proactive compaction, no context-% thresholds that trigger handoffs, and no unrequested lead or seat rotation because context is growing. Rotate or hand off only when Etan asks, or at a \u002Flarge-plan phase boundary the plan itself defines (Etan, 2026-09-24). |\n| User says \"wrap up\" \u002F \"hand off\" | Trigger handoff immediately |\n| User going to sleep with unfinished work | Handoff to continuation agent |\n| A \u002Flarge-plan phase boundary the plan defines | Handoff to the next phase's agent |\n\n**Orc seat:** orc succession uses a full-day weave (`\u002Fweave`) that seeds gen-N+1 and fires only on operator instruction. A `\u002Fcompact` from an unknown sender is an incident; investigate before complying (orc C14).\n\n---\n\n## WHAT TO INCLUDE IN HANDOFF (from \u002Ffrustration-capture)\n\nEvery handoff file MUST include the User Corrections section. This is where `\u002Ffrustration-capture` compounds:\n\n- All brain_stored corrections from this session (brain_search with \"user-correction\" tag)\n- The correction categories and importance levels\n- Behavioral rules derived from each correction\n\nWithout this, the new agent WILL repeat the same mistakes. Evidence: April 4 session — new orcClaude had no access to 34 corrections from the outgoing session because they were never stored.\n\n---\n\n## FULL-RELAY STANDARD (2026-06-05)\n\nWhen the closing window has a weave doc or retro, the handoff is NOT complete until\nthe successor has: (1) read the ENTIRE weave doc (linked, not inlined), (2) run the\ncorrections tag sweep (`brain_search`, all stored corrections of the closing\nsession) and read ALL of them, (3) ACKed item-by-item on the §EDITS and corrections\nlists. Highlights are orientation only — the dropped tail is where re-violations\ncome from (gen-11 re-violated gen-10's captured-but-unapplied edits, 2026-06-05).\n\n**Continuation\u002Flead spawns use `claude.judgment` at 1M:** in T3 through `t3_thread_launch` \u002F `delegate_task` with that model pinned; on a terminal (cmux) seat via proper repoGolem launchers.\nResolve its model with `node scripts\u002Fmodel-roles.mjs claude.judgment --field model` from\nthe golems checkout and verify the bare `{repo}Claude -s` pin against it. NEVER\npass a model param at spawn (`spawn_agent {model}` overrides the pin — 2026-06-05\nincident booted a lead at 200K). Verify `model` + `context_window` post-boot.\n\n### Gen-N successor orc seed mandate (gen-13+, operator-direct)\n\nWhen the outgoing window includes a weave doc, the successor orc boot doc MUST require:\n\n1. **Full weave-doc read** — entire doc, not highlights (`\u002Fweave` §4d).\n2. **Corrections tag sweep** — all stored corrections from the closing session, item-by-item ACK.\n3. **§EDITS application table** — state each item (APPLIED \u002F DISPATCHED \u002F TRACKED) in the successor's own words.\n4. **MONITOR + REMOTE CONTROL role** — leads **keep driving their queues**; the new orc monitors the fleet and acts as the operator's remote control, not a full re-dispatcher that respawns everything from scratch.\n\n---\n\n## ANTI-PATTERNS\n\n| Don't | Do Instead |\n|-------|-----------|\n| Send handoff as raw cmux text | Write structured file, new agent reads it |\n| Wait for user to answer grill | Outgoing agent answers from its own context (Step 4) |\n| Forget to set up monitoring | BEFORE telling user handoff is done: T3, record the successor's thread id \u002F `taskId` and confirm it is working with one `t3_thread_read`; terminal (cmux) seats only, establish `wait_for` \u002F `list_agents(detail:\"full\")` coverage |\n| Skip brain_store | Always store — file can be deleted, BrainLayer persists |\n| Start new session fresh without handoff | Even a 5-line handoff is better than cold start |\n| Declare handoff \"complete\" without verification | Check: new agent working, monitoring active, corrections stored (Step 5) |\n| Compact or hand off because context crossed a % | Let auto-compaction happen; hand off only when Etan asks or at a \u002Flarge-plan phase boundary |\n| Relay only \"the top things\" from a weave\u002Fretro | Full-relay: link the entire weave doc, mandate full read + item-by-item ACK |\n| Spawn continuation with an unpinned or wrong model | T3: `t3_thread_launch` \u002F `delegate_task` with the resolved `claude.judgment` model pinned; terminal (cmux) seats: launcher-only (`{repo}Claude -s`). Verify the model and 1M context after boot |\n| Assume a `-c` resume continues the pre-death turn's duties | Treat resume as fresh — re-derive duties from durable artifacts; encode standing orders as machinery (watch-v6 pattern) |\n\n---\n\n## INTEGRATION\n\n| Skill | How Session Handoff Uses It |\n|-------|---------------------------|\n| `\u002Ffrustration-capture` | Corrections section in handoff file — most valuable knowledge to transfer |\n| `\u002Fagent-routing` | Agent State table includes routing assignments (which agent has which workers) |\n| `\u002Forc` | W6 is the minimal version — this skill is the complete version |\n| `\u002Fcmux-agents` | Spawn-and-verify pattern from `spawn_agent` \u002F `wait_for` for the new agent |\n| `\u002Fnever-fabricate` | Verify handoff file contents before declaring complete |\n\n---\n\n### Inherited-citation suspect rule\n\nAfter consuming a session-handoff doc OR resuming post-compaction, all file:line\ncitations inherited from the prior session are SUSPECT until re-Read.\n\nThe receiving agent MUST:\n1. In Phase 0, list every inherited file:line cite with the tag\n   `[SUSPECT: inherited from {compaction|handoff}]`.\n2. Re-Read each cited path before referencing it in any deliverable. If the file\n   doesn't exist or doesn't contain the claimed content, do NOT silently drop —\n   write a \"FABRICATED IN PRIOR BRIEFING: \u003Ccite>\" entry to the deliverable AND\n   brain_store with tag:compaction-fabrication.\n3. Apply the same rule to chunk IDs (brainbar-*): re-resolve via brain_expand\n   before acting on the inherited claim.\n\nMechanism: post-compaction summaries are pattern-completed by the compacting\nmodel and have a non-zero fabrication rate.\n\n**Reason:** handoff timing estimates must use actual context telemetry:\n*\"my prior briefing had fabricated paths from a compaction summary\"* —\nreceiving agent caught fabricated paths only because the agent happened to\nverify. This rule makes it deterministic instead of relying on diligence.\n\n---\n\n### Resume-as-fresh doctrine (`-c` resumes) *(gen-13 weave E03)*\n\nA `-c` resumed session does NOT reliably execute the pre-death turn's pending\nduties — treat every resume as a FRESH session: re-derive the duty list from\ndurable artifacts (handoff file, collab, BrainLayer), never from the dying\nsession's intentions.\n\nStanding orders that must survive a restart are encoded as MACHINERY — a\nlaunchd job, or a Monitor that itself fires the action AND writes the collab\nannounce (the proven watch-v6 pattern) — not as agent to-do intentions.\n\n**Reason:** a resumed watch seat can freeze on its duties\n(ritual #2: verify\u002Fre-arm\u002Fcompletion-line never executed, ~95 min unmonitored;\nritual #3: completion line 96 min late, self-caught) [line 1538]. The fix is\nproven structurally in watch v6: *\"the monitor itself fires the ritual (nohup\ndetached) and writes the collab announce — removing me as failure point\"*\n[line 1623].\n",{"name":5,"description":7},"master","3400bc971628f00ab7bf1063993477e211a5508a","skills\u002Fgolem-powers\u002Fsession-handoff","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems","https:\u002F\u002Fgithub.com\u002FEtanHey","etanheyman.com","2026-11-15 03:05:08",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add EtanHey\u002Fgolems --skill session-handoff",{"tool":37,"command":38},"bunx","bunx skills add EtanHey\u002Fgolems --skill session-handoff",{"tool":40,"command":41},"pnpm","pnpm dlx skills add EtanHey\u002Fgolems --skill session-handoff","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fsession-handoff\u002FSKILL.md",true,[45,51],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"evals\u002Fevals.json","REFERENCE","参考文档",6899,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fsession-handoff\u002Fevals\u002Fevals.json",{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"scripts\u002Fcontext-occupancy.sh","SCRIPT","脚本",1451,"https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fsession-handoff\u002Fscripts\u002Fcontext-occupancy.sh",[58,65,72,79,86,93],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":19},"01M7C53JEF6A6SS20Q0D2YJYM8","yash","Push dependency bugs upstream. Triggers: dependency bug, workaround, monkey-patch, fork\u002Ffix. NOT app bugs.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fyash\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M7C53EPMNSZKHK47VY7WQDP0","whats-new","Diff Claude\u002FCodex\u002FCursor\u002FWispr\u002Fagy changelogs vs local setup. Triggers: whats new, changelog, version bump.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fwhats-new\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":19},"01M7C53C4TVQ9G7AB3X57DACPB","weave","Orchestrator convergence: mine JSONLs, cite findings, route actions. Triggers: weave, run weave.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Fweave\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":19},"01M7C5342EDDN7YH53KKW511V2","unslop","Cut AI tells while keeping every receipt, number, path, and hedge intact. Triggers: unslop, cut the slop, wall of text, too verbose, rambly, make this readable. NOT for delivery cadence (see i-have-adhd), channel choice, or artifact length governance.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Funslop\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":19},"01M7C52M0CTDW5DDP4452FJV0G","tmp-block","Guard denying durable writes to \u002Ftmp, \u002Fprivate\u002Ftmp, \u002Fvar\u002Ffolders, $TMPDIR; enforces \u003Crepo>\u002F.worktrees\u002F\u003Cname>. Triggers: tmp guard, TMP-BLOCK deny, WORKTREE-CONVENTION deny, bypass ledger. NOT for storage policy.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Ftmp-block\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":19},"01M7C52HN4Q9E1Z74Z11K23E2H","t3-visuals","Show\u002Fvisualize\u002Fchart\u002Fdiagram\u002Fcompare\u002Fstatus board in T3 with compact interactive HTML. Not standalone dashboards.","https:\u002F\u002Fgithub.com\u002FEtanHey\u002Fgolems\u002Fblob\u002Fmaster\u002Fskills\u002Fgolem-powers\u002Ft3-visuals\u002FSKILL.md",[],[]]