--- name: plan-blitz description: "PM-GATED. Sweep every baton in the repo that lacks an approved plan, in waves — sonnet scouts size, an Opus EM finalises, Opus planners write, plans are reviewed and integrated without the EM in the loop, and the EM gates readiness at the end. Or target named batons. Wave N+1 fires on wave N's approvals, not its landings." description-budget: 320 version: 1.0.0 allowed-tools: ["Read", "Write", "Edit", "Bash", "Grep", "Glob", "Agent", "Skill", "Workflow", "AskUserQuestion", "TaskCreate", "TaskUpdate", "TaskGet", "TaskList"] argument-hint: "[ ...] [--roadmap-id ] [--waves ] [--dry-run] [--repair ...]" --- # Plan-blitz — a roadmap's worth of plans, in waves Consumes a roadmap that already exists and produces plans for it, N batons per wave instead of one baton per session. Argument and worked rationale live in the fleet doctrine wiki under this skill's own name — read it when a rule here looks wrong, never to decide whether to follow one. The two tripwires below are the greppable entry points. --- ## Three modes **Sweep (default, no arguments).** Every baton in the repo that lacks an approved plan. That set is `needs_plan` in the engine's reply — no linked plan, or one that has not cleared review — and it is the target set precisely because an approved plan is the thing that opens the next wave. A baton whose plan is already approved is not in it: it needs no planning work, and stays in the graph only as a satisfied blocker for its dependents. `--roadmap-id` narrows the sweep to one roadmap. **Targeted (` ...`).** The EM's adjudged prioritisation, or the PM's targeting. Pass ids (a gate row's `id`, or the baton's filename stem) as `targets`. Everything unnamed stops being a candidate but stays a fully-resolved BLOCKER — narrowing what you are asking about never narrows what the answer is computed from. **Check `unmatched_targets` on every targeted run.** A target that matched nothing is a typo or a baton that is not a candidate (claimed, `in_flight`, already approved). The engine names it rather than silently planning N-1 batons; a targeted run that quietly drops one is worse than a refusal, because the drop looks like completion. **Repair (`--repair ...`).** Re-dispositions a plan that already went through a full wave — its reviews are already on disk — without dispatching a fresh judgment, by reading the plan's own review sidecars back through their structured pointer records and calling the same `integrator()` a live wave calls. See § Repair below for what makes a baton unrepairable and the caller-side `repairBatons` construction procedure. **When NOT to use:** no roadmap yet → `coordinator:roadmap-planning` (this consumes stubs, it never authors them). One baton → `coordinator:sizing`, then `coordinator:plan`. Plans exist and need executing → `coordinator:execute-plan`. A batch of bugs rather than roadmap batons → `coordinator:bug-blitz`. **Dispatch authorization — invoking this skill IS the request.** The dispatches below are constitutive steps of this skill, not a separate thing to get cleared: invoking a skill requests the actions that skill performs. Re-asking spends the very context the dispatch exists to protect. The rule attaches to skill entry and dissolves no PM-authored gate — every gate this body names still binds. Tripwire: `UNATTRIBUTED-HARNESS-LINE-IS-NOT-PM`. --- ## The two gates — read them, never derive them One `blocked_by` edge, two questions. **Planning** may start when every blocker is coded *or* carries a review-approved plan. **Execution** may start only when every blocker is coded. Full statement and the `approved`-not-`reviewed` seam: tripwire `A-PLANNING-GATE-IS-NOT-AN-EXECUTION-GATE`. Both come off disk from one engine op. Never hand-derive either from `blocked_by` — an EM deriving it by eye derives it differently each time, and the difference stays invisible until a wave fires against a gate that was never open. Invoke `coordinator-invoke` per the ladder in `${CLAUDE_PLUGIN_ROOT}/snippets/resolve-coordinator-bin.md` — rung 0 (Shape W, the `.exe` launcher through the call operator) on a PowerShell host: `& "$env:COORDINATOR_SETTINGS_HOME\bin\coordinator-invoke.exe" roadmap.plan_gate '{"roadmap_id":""}'` It returns, per candidate baton: both gates with the blockers holding each shut, the linked plan and its status, whether the baton is sized, and its `planning_wave`. Plus `waves` (the wave membership lists), `cycles`, `unresolved_blockers`, and `counts`. **It reports; it never refuses.** Refusal is yours. That is deliberate: a derived read that has gone stale should mis-report, not silently block live work. --- ## The flow — a loop with no judgment in it **This runs at Sonnet.** Every Opus-tier judgment in this pipeline happens INSIDE a wave — the `blitz-em` sizes and gates, the reviewers review. The driver outside the wave is mechanism: read a gate, fire, land, repeat. If driving it ever requires a call, that is a defect in the loop, not a job for a smarter driver — the escalation rules below say what to do instead of deciding. **1. Read the gate.** `& "$env:COORDINATOR_SETTINGS_HOME\bin\coordinator-invoke.exe" roadmap.plan_gate '{}'` Three fields before anything else, each with a mechanical response: | Field | If non-empty | Response | |---|---|---| | `unresolved_blockers` | an edge names no record on disk | **Read the `baton` column first** — see below. Stop only for a member of the wave you are about to fire. | | `cycles` | batons block each other | **Stop and report** the named members. | | `counts.unschedulable` | blockers this pass cannot clear | Proceed; they are excluded by design. | **An unresolved blocker is scoped to its own baton, not to the run.** The gate fails it CLOSED — both gates shut, the baton out of every wave — so a sweep over the other batons is planning against no unread gate. Stop only when a named `baton` is a member of the wave you are about to fire; otherwise report the entries and proceed. Halting a 40-baton sweep over a defect on a baton the engine already excluded is the more expensive error, and it recurs on every later invocation. **A blocker naming a peer EM (`doe-claude-em`, `claude-klabauter-em`) is the standing case, and it is not a typo to repair.** `blocked_by` takes stub ids and `handoff_id`s, so a role name resolves to no record and lands here permanently — the shape a baton waiting on a cross-repo answer wears. **Never clear one by deleting the edge**: the edge is the only thing holding both gates shut, and removing it makes the baton a live planning candidate the moment the sweep re-reads. It clears when the peer answers. **An edge is a DISCOVERED dependency, never a way to suppress a candidate.** A peer EM owes an answer, which is why that case is an edge. A baton you have decided not to fire — a PM ruling, an adjudication, a box in the wrong state, a wait on a peer's landing — owes nothing to anybody, and spelling it as an edge the gate cannot resolve fails the baton CLOSED forever — worse than the retype it was meant to save. That is a standing ruling, restated in the engine's own gate and in `handoff.schema.json`'s description of the hold fields below. **The mechanism for that is a HOLD, three flat baton keys the gate reads: `plan_blitz_hold_reason`, `plan_blitz_hold_cite`, `plan_blitz_hold_until`.** `roadmap.plan_gate` sets `candidate: false` and reports the baton under `held` with its reason. **A hold with no reason is not honoured** — an unexplained suppression is the thing this replaces, not a lighter version of it. `until:` is where a real expiry goes (a peer's re-vendor, a publish, a named event). **A hold is what `--exclude` was standing in for.** A per-fire exclusion list is retyped every run, lives in one driver's head, and is invisible to the next session — measured on Example-game-workbench-repo, 13 batons held that way across three waves by an EM who did not know the field existed. `--exclude` narrows one fire; a hold is the durable record. Reach for the flag only when a wave is already part-fired. Holds cover more than they look like they do: a PM ruling, two batons sharing one plan, a replanned source, a repair-lane baton, an adjudication standing three waves, a grouping assent, a live-editor-session or clean-machine precondition, and a baton waiting on a sibling repo's landing. **No separate host or precondition vocabulary is minted for these** — a reason in prose a reader can evaluate beats a token for a case that occurs twice, and a `host:` token would be inert AND false on a box that matches on platform but is in the wrong state. **What does NOT work, and reads as though it does: `external_gate` on the baton record.** `roadmap.plan_gate` never reads the key — candidacy is `live`, `baton_role`, `status` and `deployment_state` — and its scanner declares every baton key it reads, so a key outside that set reads as absent. `external_gate` is real, but it belongs to PLAN SPINE ROWS, where it withholds rows inside a plan that still certifies. Written onto a baton it is well-formed frontmatter that changes nothing, and the next wave's EM reads it and concludes the question is settled — worse than the open defect. Reported by example-game-workbench-repo-b8 after two independent wave EMs recommended it; four of their batons were being held out of waves by hand-maintained exclusion lists for want of the edge above. **2. Scaffold the trail and freeze the gate.** `state/plan-blitz//wave-.gate-report.json`, where `` is this run's UTC start as `YYYYMMDDTHHMMSSZ` — every trail on disk is one, and `recycle-check` orders runs by it. **One report PER WAVE, named for its wave — a run-wide `gate-report.json` is wave 0's name only.** Every wave freezes its own, after the previous wave landed. A report frozen for an earlier wave is SHAPE-RIGHT and TIME-WRONG: every `batons[]` lookup resolves, nothing refuses, and the manifest reads as a healthy wave several times the real one's size, carrying plans the run has since approved and plans open in a live agent. That is the opposite failure to `AN-ENVELOPE-FROZEN-AS-A-GATE-REPORT-READS-AS-AN-EMPTY-WAVE`, which fails loudly and empty. `emit-wave-fire.py` refuses it against the previous wave's landing declaration; pass the path and the refusal never has to fire. The workflow has no filesystem primitive — an unscaffolded directory means every sidecar write lands nowhere and the readiness gate reads an empty trail. **Pass `--bare`.** A successful invoke then prints `result` alone — the documented shape — and an error still prints the whole envelope, so a failure is never frozen as a valid-looking report. Everything below is what you get WITHOUT it. **Redirect `coordinator-invoke`'s stdout and what you froze is a JSON-RPC envelope** — the gate's own `waves`, `batons` and `counts` sit one level down, under `result`. That is the shape the readers below accept, and the shape to build the baton array from. A machine reader that asks the envelope for `waves` finds none and reports an EMPTY WAVE, which is the same sentence a genuinely empty wave produces — so the wave fires with step 2a skipped and nothing says so. Tripwire: `AN-ENVELOPE-FROZEN-AS-A-GATE-REPORT-READS-AS-AN-EMPTY-WAVE`. Redirect stderr separately: the engine writes a `[warm-settings]` line there, and `2>&1` folds it into the JSON. **The readers tolerate either shape anyway** — `recycle-check.py` and `emit-wave-fire.py` both unwrap an envelope — but freeze the documented shape rather than relying on that: a consumer written later will not. **Freeze by redirection** (`1> gate-report.json 2> gate.stderr`); `ConvertFrom-Json | ConvertTo-Json` truncates every baton row at its default depth. **Redirection also decides ENCODING, and PowerShell 5.1 and 7 disagree** — 5.1 writes a UTF-8 BOM (`>`/`Out-File` there default to UTF-16LE), pwsh 7 does not. A BOM fails a plain `json.load` downstream and reports as a malformed gate. Read back with `utf-8-sig`, correct on both. **2a. Check the wave for finished work.** One pure read, before any agent is dispatched: python3 "${CLAUDE_PLUGIN_ROOT}/skills/plan-blitz/recycle-check.py" --repo-root \ --gate-report --exclude-run `RECYCLED` (exit 1) names a baton whose execution record from an earlier wave says the work FINISHED, and which the gate still returns as a candidate — its landing never stamped it, because `blitz_land` refuses the XS lane without `shipped_in`. The repair is that landing, re-run with the SHA; **never drop the baton from this wave by hand**, which leaves it to recycle into the next one. `RESURRECTED` (exit 1) is a candidate already archived terminal: remove or re-archive the live copy. `back` entries are `completed: false` or `blocked-on-preflight` and are correctly here. Tripwire: `A-FINISHED-BATON-THE-LANDING-NEVER-STAMPED-COMES-BACK-AS-A-CANDIDATE`. **3. Fire the wave. Emit it; never hand-write the args.** python3 "${CLAUDE_PLUGIN_ROOT}/skills/plan-blitz/emit-wave-fire.py" \ --repo-root --trail-dir --wave-index N --wave-number N \ --gate-report /wave-N.gate-report.json It reads the frozen gate report, derives every per-baton field from it, splits the wave into fires at the cap, binds each fire's args into a standalone `.mjs` through claude-klabauter's `workflow.bind_args`, and prints one `Workflow({ scriptPath })` line per fire — **with no `args` at all**, because they are bound into the file. Fire each printed line. Pass `--plugin-agents-available true` where `coordinator:*` types resolve. **An emitted fire is a SNAPSHOT, never a handle — RE-EMIT, never re-fire as found.** It binds this host's paths, so one from another box is wrong in space, and the workflow BODY as it stood at emit time, so one from this morning is wrong in time. The second is the trap: it runs and succeeds. Measured — a repair re-binding the args of an 11:20 fire reproduced a contract defect fixed at 12:17, costing its driver an afternoon re-deriving rulings the source already honoured. **A hand-typed args object is the defect this replaces, not a shortcut past it.** Everything the args contract can lose, it loses silently: `executionOpen` reached a live wave missing because a caller copied an example that omitted it, and the wave then dispatched no XS, closed none at the landing, and reported them under `routedElsewhere`. A derived field cannot be omitted. An emitted script is also on disk, so it archives with the trail and the fire can be re-read, re-fired and diffed — an args object inside a tool call is none of those. Tripwire: `A-HAND-TYPED-WAVE-ARG-IS-AN-UNARCHIVED-FIRE`. **A fire is a SNAPSHOT of the workflow, so a fix to `plan-blitz.mjs` never reaches a script already emitted.** Re-firing or resuming an old one runs the old body against today's content, and such a bug fires on CONTENT, not on age: in one measured run, 23 fires each carried the same unguarded field access, and only the one whose integration raised the triggering shape died — after 40 agents had completed. **Before re-firing or resuming a script emitted before your last workflow fix, diff it against the live `plan-blitz.mjs` or re-emit it.** Do not retro-patch a fire that already ran: the archive is what ran, and rewriting it makes the trail lie. **Patching a script to resume it re-runs every agent DOWNSTREAM of the text you edited, not none of them.** Resume caches on `(prompt, opts)`, and the prompts interpolate the workflow's own helpers — so editing a function a later phase's prompt calls changes that phase's cache key and it runs live. Measured: a fire patched at its assembly step replayed 21 of 40 agents, the whole tail. That is the right outcome and cheap for a fire that died at the END; a patch further upstream re-runs almost everything, and re-emitting is then the cheaper repair. Decide by where the edit sits in the phase order, never by the size of the diff. **Pass `--wave-number` with the run's own wave count.** `roadmap.plan_gate` numbers `waves` from the READ, so every wave arrives as `waves[0]`: `--wave-index` selects which wave of the frozen report to fire and stays 0, while `--wave-number` names the bound `waveIndex`, the fire filename and the trail slot. Omit it on a one-wave run. Never hand-edit the frozen report to renumber. Batons come from `waves[N]`, **at most 8 per fire** (§ batching above). A wave larger than 8 is drained by several fires at the same `waveIndex`, sharing one trail directory. That is supported: the wave-scoped sidecar is keyed by the fire's own baton set, so fires do not overwrite each other's size review. The emitter does that split; do not renumber the wave to separate them — `waveIndex` is what the gate computed, not a fire counter. **`--exclude ` narrows what is FIRED and never what the gate computed** — its case is a wave already part-fired, where re-emitting a live baton would run two waves against one plan. **Every fire writes its records into its OWN leaf of that trail**, `/wave--/`, returned as `trailSlotDir`. So neither a second fire of one wave nor a LATER wave re-planning a baton this one already planned overwrites what is on disk — search the trail RECURSIVELY, and read a second slot holding the same baton as that baton's record from a DIFFERENT fire rather than a duplicate to reconcile. **Order slots by INTEGER wave index, never lexically** — `wave-10-…` sorts before `wave-2-…` as text, so a ten-wave run's oldest record reads as its newest; `recycle-check.py :: _slot_order` is that ordering, do not re-derive it. The args contract itself lives at the top of `workflows/plan-blitz.mjs` and stays the source of truth for what the emitter produces. Read it to understand a field; do not assemble it by hand to fire one. **Fires may run CONCURRENTLY, and the driver owns disjointness — the gate cannot.** Concurrency is what makes a 200-baton wave finish, since a fire costs ~50 minutes of wall clock whatever its size. But `roadmap.plan_gate` reports the batons that need planning, and a baton being planned *right now* still needs planning: nothing on disk changes until that fire lands. So a driver re-reading the gate gets its own in-flight batons back at the head of the list, and firing them plans one baton twice — two waves writing one plan file, and whichever lands second overwrites the first. **Subtract your own in-flight set from every fresh gate read.** Keep it in the run's own notes; the engine has no session-scoped view, and adding one would make a derived read authoritative over disk. **Settling a pull edits a plan the next planner cannot see.** Adjudicating a pulled baton means editing its plan — but a planner reads the BATON, not the trail and not your settlement, so re-firing it re-authors the plan over your ruling, silently. **Re-freeze the gate before re-emitting a baton whose plan you edited**; the emitter warns when a plan changed after the frozen report. **A hold that only you remember is not a hold.** In-flight ends with the run, but a standing reason not to fire — adjudicated `pulled`, a DR that settled it, a blocker with no edge — outlives the session that learned it, and the next gate read hands the baton straight back. Measured on project-rag: 5 of 8 wave-0 candidates had a recorded reason not to fire, four re-fired across five runs. Write it where the next driver reads it: the trail's `RUN-NOTES.md`, and the baton record itself when the reason is the baton's. A note still has to be FOUND, so prefer the baton record whenever the reason belongs to the baton — a gate read sees that; it does not see your notes. **Disjointness is on `planPath`, not just on baton id — this is the one that bites.** Several batons routinely share one plan, and a roadmap plan links every baton it emitted: NINE share one here. Two fires holding different batons that name the same plan are two waves authoring one file, and whichever integrator writes last wins, silently. The sidecar keying above does not help — it makes each fire's SIZE REVIEW safe and says nothing about the plan. Same plan, same fire, or fires that do not overlap in time. **`emit-wave-fire.py._split` already enforces that WITHIN an emit — do not hand-audit it.** Your residual duty is ACROSS emits: a fresh gate read cannot see that a live fire holds a baton naming the same plan, so subtract in-flight `planPath`s before the next emit. **Subtract adjudicated batons too, for the same fire.** A `pulled` baton stays a candidate by design — the EM left it where it was — so it also returns at the head of the next read. Re-firing one immediately re-runs the wave that just judged it, against an escalation nobody has answered in between. Fire it again when its pull reason is addressed, not because the gate still lists it. Tripwire: `A-GATE-READ-DOES-NOT-KNOW-A-FIRE-IS-RUNNING`. **`dispositionsCli` and `provisionSidecarCli` are resolved caller-side, and `emit-wave-fire.py` does it** — launcher first, then the `--engine-root` checkout's own `coordinator/bin/`, with the interpreter and any env the registry manifest needs made part of the injected literal. Pass `--dispositions-cli` / `--provision-sidecar-cli` only to OVERRIDE that, never as the ordinary path. Omit them on a box with no install and a reviewer invents its own sidecar path: the review runs, the disposition record is lost, and nothing reports it. **A resolved `provision-sidecar` whose findings path carries no `subagent-share` segment is a REFUSAL too** — the disposition op rejects every target without it, so the wave reviews and records nothing. Tripwire: `A-SIDECAR-THE-DISPOSITION-OP-REFUSES-LOSES-ONLY-THE-RECORD`. **A box with no install is Rung N, not a broken resolution.** Every launcher rung fails command-not-found and reads as "this CLI does not exist"; the engine source is on disk regardless, so pass `--engine-root`. Where the only checkout carrying `workflow.bind_args` is claude-klabauter's unstamped authoring tree, add `--live-engine-tree` — it takes the engine's live-tree path for the BIND CALL alone, and refuses a root that carries a build stamp. Tripwire: `A-PUBLISHED-MIRROR-OLDER-THAN-AN-OP-REFUSES-AS-A-MISSING-ENGINE`. **Every baton carries `executionOpen`, read off that baton's own `execution_gate.open` in the gate report.** It is not optional and it has no default: an XS is dispatchable only when its EXECUTION gate is open, so a baton missing the field fails that test, dispatches nothing, and is never closed at the landing — it comes back as a candidate in every later wave. The wave reports it under `routedElsewhere` either way, which is why the omission is silent. The full per-baton shape is the args contract at the top of `workflows/plan-blitz.mjs`; build the array from the frozen gate report, never by hand — `emit-wave-fire.py` derives it, and refuses rather than defaults a baton row carrying no `execution_gate.open`, because a field whose wrong value costs a whole wave silently cannot have a default. Resolve `${CLAUDE_PLUGIN_ROOT}` — do not pass a repo-relative path. The plugin root differs by tree: under the DoE source repo it is the `coordinator/` subdirectory, and in an installed or mirrored plugin it IS the root. A path written `coordinator/workflows/...` resolves only when the cwd happens to be DoE, and elsewhere fails as a MISSING FILE, which reads as "the vehicle does not exist" rather than "the path was not resolved" — the more expensive of the two wrong conclusions. Then wait. **Do not read the trail to decide anything** — reading it to follow along is fine and costs nothing, but the wave needs no input between fire and return. A driver that intervenes mid-wave is overriding a judgment the `blitz-em` was dispatched to make. **4. Commit the wave's XS work, then land it — one call over every fire of the wave.** python3 "${CLAUDE_PLUGIN_ROOT}/skills/plan-blitz/land-wave.py" \ --repo-root [--shipped-in ] ... One file per fire — the harness's task-output file the workflow's completion named, or the bare result pulled out of it; both are read. It lands each fire through `roadmap.blitz_land`, **sums the three lanes across every fire**, and states the stop condition from that sum. It refuses before landing anything if the fires disagree about `waveIndex`, or if any fire dispatched XS and no `--shipped-in` was given. Read § the op below for what the landing DOES — the helper orders the calls and does the arithmetic; it decides nothing, fires nothing, and re-queues nothing. It writes the wave's own `wave-.landing.json` into the trail, which is what the NEXT session reads instead of your stdout. **Never redirect its stdout onto that path** — the text report and the script's own write interleave into a file no reader can parse. Capture stdout elsewhere if you need it. **Commit before landing whenever the wave dispatched any XS.** `close_dispatched` stamps the baton `shipped` with a `shipped_in` SHA, and this op does not commit — a stamp written first would cite a commit that does not exist. Pass that SHA as `shipped_in`; without it the XS lane refuses and those batons stay open: the recycling defect, not a cosmetic gap. `& "$env:COORDINATOR_SETTINGS_HOME\bin\coordinator-invoke.exe" roadmap.blitz_land '{"wave_result": , "shipped_in": ""}'` `roadmap.blitz_land` executes the verdicts the readiness gate already made: it links each `ready` plan to its baton and *then* stamps it `approved`, mints a baton per `replan` carrying the brief verbatim, leaves `pulled` where the EM left it, and returns `next_wave` computed from a **fresh** gate read taken after the writes. **Three lanes land differently, and the op picks by route — you do not.** | Route | Size | What landing does | |---|---|---| | `plan` | M / L | link the plan to its baton, then stamp it `approved` — this is what opens the next wave's planning gates | | `spec-dispatch` | S | park the spec onto the baton and stamp it execution-ready (four-field `execution_authorized_*` + `handoff_phase: execution`), so `/execute-plan` resolves it as a straight dispatch — mise-en-place tier | | `dispatch` | XS | work already done inside the wave's Dispatch phase; landing stamps the baton `shipped` with `shipped_in`, which is what makes it terminal and stops it returning as a candidate | **Why S parks rather than approves.** An S is a straight dispatch, not decision-weight work. Sending one round a full review cycle and then handing back an un-actioned baton is what made the EM's own sizing fight the rest of this skill: if calling something S condemned it to the queue, the honest S got inflated to M. Sizing that bends toward its downstream route is corrupted sizing. An S baton's plan stays at `draft` by design — `needs_plan` keys off the execution stamp for these, so a later sweep does not re-plan work that already has its marching orders. **Never hand-stamp `status: approved` instead.** The op links before it stamps and refuses to stamp what it cannot link: an approval resolving to no baton is a silent no-op that reads as success, measured twice here. A hand stamp skips the check that exists for that. **Read `refused[]` on every landing.** Each entry names a baton and why. A refusal is the op declining to write something misleading; it is never a thing to route around. **5. Loop.** Fire `next_wave` and land it. Repeat. **Stop conditions, all mechanical:** - `next_wave.batons` is empty — nothing left to plan. - `--waves` exhausted. - A wave lands zero `approved` **and** zero `execution_ready` **and** zero `closed` — it opened nothing, so the next wave is this wave again. Report and stop rather than spinning. **Zero `approved` alone is not the test.** A wave whose ready set is all S routes approves nothing by construction and still advances: `blitz_land` parks each S execution-ready, and `needs_plan` keys off that stamp (§ Three lanes above says so in as many words). Reading this condition as `approved == 0` halts a healthy blitz and reports it as a stall. Measured 2026-09-10 on project-rag, one fire of a five-fire wave: `approved: 0`, `execution_ready: 2`, and a candidate set that still falls — `needs_plan` 27 to 24, `remaining` 18 to 15. Tripwire: `AN-ALL-S-WAVE-APPROVES-NOTHING-AND-STILL-ADVANCES`. **And it is a WAVE-level test, not a fire-level one.** A wave over 8 is drained by several fires sharing one `waveIndex` (§ batching), but `blitz_land` takes ONE fire's result, so the landing you are holding is a fraction of the wave. Sum the three lanes across every fire at this `waveIndex` before concluding anything. Same wave, same day: one fire of five landed all three lanes at 0 — genuinely nothing — while the wave around it had opened 9. **And the test only applies to a wave that FINISHED.** A fire killed part-way — a rate limit, a crashed host, a cancelled run — returns the identical zero-lane signature: it never reached its readiness gate, so it has nothing to report in any lane. "It never ran" and "it opened nothing" want opposite responses, stop versus resume, so check completion BEFORE reading the lanes. The tells are in the result and its diagnostics, not the counts: agents that errored, and every `converged` row reading `skipped: "no integration report to resolve over"`. A journal's `{"type":"failed"}` record carries no label, phase or error text — join its `agentId` back to its own `started` line, and read the cause off that agent's transcript's last message. **Never land an unfinished fire** — its `ready: []` is an absence of judgment, not a judgment of nothing, and stamping it writes that absence to disk as a verdict the wave never made. Resume instead: a run replays every agent that already completed, so recovery costs only the ones that died. **A run id resumes only in its own session**; after that, re-emit and re-fire, and close a finished XS with `archive-stamp-cli ship-handoff --sha `. Measured 2026-09-10: four concurrent fires hit one account session limit within minutes and all four returned `ready: []`. An 8-baton Opus fire is close to a limit-length run by itself. **The ceiling is the ACCOUNT, not the box** — 2026-09-11, three fires here and a fourth in a peer repo died in one minute on one limit. Budget against what peers are running, not your own fire count. Tripwire: `AN-UNFINISHED-WAVE-IS-NOT-A-WAVE-THAT-OPENED-NOTHING`. - `refused[]` is non-empty — report and stop; a landing that could not complete must not be built on. - `surfacedToPm` is non-empty — those need a PM answer. Carry them out; **never re-queue one.** **What the driver escalates rather than decides:** an unresolved blocker, a cycle, any `refused` entry, anything in `surfacedToPm`, and a wave that lands nothing. That is the complete list. Every other outcome has a defined next call. --- ## Rules that hold across every wave **Review fires unconditionally; the EM gates once, at the end.** No mid-wave permission to review, no per-plan "does this need the Staff Engineer?" Gating review on the EM makes addressing a finding cost effort and ignoring it cost nothing; firing it by default inverts that, so declining a finding becomes the deliberate act. The EM's authority is not reduced — it moves to reading a durable trail. Tripwire: `A-BLITZ-WAVE-THAT-GATES-ON-THE-EM-IS-NOT-A-BLITZ`. **Host availability on the box the wave runs on is never a pull reason.** The readiness gate asks whether the plan can be RUN — by whoever runs it, on the host it names — not whether it can be run here, now, by the gate. A plan whose rows are withheld behind a declared `external_gate` for a host this box is not is READY, and the withheld rows are a schedule fact, the same way a non-empty `mise_prepped_findings` is one. Pull for properties of the PLAN: an unapplied finding, an unsettled escalation whose answer changes the deliverable, self-contradictory acceptance criteria. Tripwires: `A-PLANNING-GATE-IS-NOT-AN-EXECUTION-GATE`, `THE-BOX-THE-WAVE-RAN-ON-IS-NOT-THE-BOX-THE-PLAN-RUNS-ON`. **One reviewer-attributed option is a recommendation, not a choice and not a dead end.** Two or more contested options are arbitrated by the resolve pass, which picks one; exactly one is applied or declined with a reason, which is ordinary integration work and not arbitration. Neither lane may carry an option the integrator composed — the attribution filter drops those before either lane sees them, and that bound is what the two-option floor protects. A recommendation the pass neither applied nor declined reconciles a `ready` verdict to `pulled` mechanically, so something decides every one of them. Tripwire: `A-SINGLE-REVIEWER-OPTION-IS-A-RECOMMENDATION-NOT-A-DEAD-END`. **No reviewer is prescribed in a plan file.** Reviewers are resolved per baton by the blitz-em from what that plan actually needs. A reviewer named on every plan is a reviewer nobody chose. **`BLOCKED` and `PIVOT` are different questions, not a severity ladder.** `BLOCKED` means "wrong until you fix these" — the direction holds, the integrator applies the findings, and the fixed plan can be approved in the same wave. `PIVOT` means "this direction cannot proceed" — no set of findings repairs it. The test that separates them: if you can name what a competent author changes in this plan to make it right, it is `BLOCKED`, however large that change is. Reviewers picking by "how bad is it" pick wrong, which is why `PIVOT` is reached by judging direction and must carry a stated premise failure. Tripwire: `A-BLOCKED-REVIEW-IS-NOT-A-PIVOT`. **`PIVOT` routes; it does not halt.** A reviewer rejecting a plan's premise has produced exactly the evidence a replan needs. The wave records it, turns it into a replan baton, and the other N-1 plans finish. Killing the wave to report one pivot discards the work that succeeded. **Every reviewer's verdict survives, separately.** Sidecar paths are assigned per reviewer, a mixed set is named in the trail, and a pivot's suspension of the plan does not delete the co-reviewers' findings — they are triaged as `Suspended (PIVOT)` and carried into the replan brief. Two reviewers choosing the same obvious filename cost a whole BLOCKED review once; assignment is why they cannot again. **Never blitz a claimed or `in_flight` baton.** They have a live holder, and a wave writing a plan for work somebody is already doing races them. The op excludes them from the candidate set; do not add them back by hand. **Never open a gate the engine says is shut.** If `roadmap.plan_gate` reports a planning gate shut, it is shut. The repair is to fix the blocking edge or clear the blocker. **Never let the blitz-em resolve a PM decision.** `route: pm-decision` and XL exits leave the wave in `surfacedToPm`. The blitz-em is an EM proxy, never a PM proxy. **`approved` is not `mise-prepped`, and a wave never stamps one.** Landing opens the *planning* gate; the `mise_prepped_by/_at/_sha/_findings` attest says a hands-off run may fire the plan without an overseer, and it is stamped by `plan.stamp_prepped` outside this skill, driven by `/mise-prep`. This skill stops at *ready to execute* in both vocabularies. **A wave's body edits invalidate a stamp; its frontmatter writes do not.** `mise_prepped_sha` is `canonical_body_sha` of the plan BODY, frontmatter excluded — so `blitz_land`'s `status: approved` flip, its baton link and its `execution_authorized_*` park all leave an existing stamp intact by construction. The review-integrator applying findings rewrites the body, and that does invalidate it: the plan is then STALE, and STALE re-gates rather than re-stamps. Never read `mise_prepped_by` for presence. Tripwire: `A-PRESENT-MISE-PREPPED-STAMP-IS-NOT-A-CERTIFICATION`; consumer contract: `coordinator/docs/wiki/mise-prepped-attest.md`. --- ## Repair **Emit it; never hand-build `repairBatons`.** The rule from § Fire the wave holds here for the same reasons — an args object inside a tool call is not on disk, so the repair cannot be re-read, re-fired or diffed, and nothing archives with the trail: python3 "${CLAUDE_PLUGIN_ROOT}/skills/plan-blitz/emit-wave-fire.py" \ --repo-root --trail-dir \ --repair [--repair ...] It resolves each baton's pointer records itself under the rules below, refuses per baton with the reason on stderr, and writes `repair-fire-.mjs` into the trail — numbered past any repair already there. Fire the printed `Workflow({ scriptPath })`, with no args. No gate report is read. The bound shape, for reading a fire rather than writing one — one entry per plan re-dispositioned: ``` { batonId, planPath, reviews: [ , ... ], unresolvedPointers: [ ... ] } ``` **Where the records come from.** A wave has each reviewer provision its findings sidecar through the `provisionSidecarCli` its caller resolved, and leaves in the trail only a pointer record naming it: `{ sidecarPath, verdict, premiseFailure }`. The emitter resolves each pointer from the latest WAVE slot only, and skips `repair-/` slots — a repair re-emits no reviewer, so its slot holds an integration report and no pointers, and reading one hands `reviews: []` to a baton whose reviews are one directory over. **Latest is by INTEGER wave index, never lexical** — `wave-10-…` sorts before `wave-2-…` as text; `recycle-check.py :: _slot_order` is the ordering, imported rather than re-derived. A record whose `sidecarPath` still exists is a review; one whose target is gone is an `unresolvedPointer`, and any of those refuses the whole baton. **Refusals, all loud, none silent.** The whole baton or none of it — never a subset — when: no `planPath`; any `unresolvedPointers`; an empty `reviews`; any record missing `verdict`; or no `trailDir`. Every refusal names the baton and why, because a repair that quietly clears a plan it found nothing to disposition reads identical to one that re-dispositioned it. A missing `verdict` is the pre-contract bare-path shape, so a trail older than the pointer record is unrepairable: repair does not rescue the backlog that motivated it. **A pointer record names its plan by path, never by version, so check the plan has not moved under the review** — `git log -- ` against the sidecar's mtime. A rewrite between review and repair makes the findings stale, not wrong, and the repair re-escalates fixes the plan already carries. Tripwire: `A-REVIEW-SIDECAR-NAMES-ITS-PLAN-BY-PATH-NOT-BY-VERSION`. **A repair produces an integration, never a verdict — so a repaired plan is not yet `approved`.** There is no landing for a repair result: `blitz_land` lands a WAVE. The route to approval is one targeted re-fire of that baton, and it is safe because the emitter binds the plan this trail already holds even when the gate does not link it (a pulled baton comes back unlinked), so the planner REVISES rather than authors over your settlement. Re-freeze the gate first if you edited the plan. Do not hand-stamp `approved` — § the op says why. **What it does not do.** It dispatches no `sizingScout`, no `planner`, no `reviewerAgent`. The one role it reaches is `integrator()`, unforked, holding no `Agent`/`Task` tool and so unable to spawn one transitively. Verdicts resolve through the same `resolveVerdict()` a live wave uses, so a premise failure becomes PIVOT here exactly as it would in a wave. --- ## Anti-scope - Does not author roadmap batons (`coordinator:roadmap-planning`). - Does not execute plans, and never opens an execution gate. It stops at *ready to execute*. **That boundary is not conditional on who reads the exit.** The landing is durable on disk (`status: approved`, `governing_plan`), so mise-prep's entry reads it back through `roadmap.plan_gate` — a read reports what a gate would say and never opens one. Consumer: `skills/plan-blitz/mise-prep-entry.py`. Tripwire: `A-HANDOFF-AN-EM-RETYPES-IS-NOT-A-SEAM`. - Does not ratify sizings on the PM's behalf. - Does not review code. The reviewers in a wave review **plans**. --- ## Test Surface No runtime test for this skill body — prose doctrine, not executable code. The executable surface is in claude-klabauter (`coordinator_core/roadmap/tests/test_plan_gate.py`, `coordinator_core/ops/tests/test_roadmap_plan_gate.py`) and in this repo's `coordinator/tests/test_plan_blitz_contract.py`, which asserts the workflow script conforms to the Workflow correctness contract and that the doctrine below stays greppable. | # | token | file | expect | threshold reason | |---|---|---|---|---| | 1 | `A-PLANNING-GATE-IS-NOT-AN-EXECUTION-GATE` | this file + `skills/pickup/SKILL.md` + `skills/execute-plan/SKILL.md` | ≥3 | the two-gate rule is inert if only the skill that introduced it knows about it | | 2 | `roadmap.plan_gate` | this file + both consuming skills | ≥3 | a gate nobody reads is a gate nobody honours | | 3 | `A-BLITZ-WAVE-THAT-GATES-ON-THE-EM-IS-NOT-A-BLITZ` | this file + `agents/blitz-em.md` | ≥2 | 1 means only the self-reference survives | | 4 | `plan-blitz.mjs` | this file | ≥1 | the vehicle is named, not left to be rediscovered | | 5 | `A-PRESENT-MISE-PREPPED-STAMP-IS-NOT-A-CERTIFICATION` | this file + `skills/review/SKILL.md` | ≥2 | 1 means only the surface that introduced it knows the predicate is a recomputed sha | | 6 | `THE-BOX-THE-WAVE-RAN-ON-IS-NOT-THE-BOX-THE-PLAN-RUNS-ON` | this file + `agents/blitz-em.md` | ≥2 | the surface that PULLS has to hold the rule; 1 means only the skill describing the gate knows it | | 7 | `A-SINGLE-REVIEWER-OPTION-IS-A-RECOMMENDATION-NOT-A-DEAD-END` | this file + `agents/blitz-em.md` | ≥2 | the gate reads APPLIED/DECLINED lines it has no vocabulary for otherwise | | 8 | `AN-ENVELOPE-FROZEN-AS-A-GATE-REPORT-READS-AS-AN-EMPTY-WAVE` | this file + `skills/plan-blitz/recycle-check.py` | ≥2 | the reader that unwraps and the step that tells you to freeze must both carry it; 1 means the code is tolerant and the instruction still teaches the trap |