tickmarkr 1.74.0 → 1.75.0

package/dist/adapters/claude-code.d.ts

@@ -1,4 +1,5 @@
-import { type AuthHealth, type WorkerAdapter } from "./types.js";
+import { type AuthHealth, type TrustDialog, type WorkerAdapter } from "./types.js";
+export declare const CLAUDE_TRUST_DIALOG: TrustDialog;
 export declare function claudeSlug(real: string): string;
 export declare function probeVersion(bin: string): AuthHealth;
 export declare const claudeCode: WorkerAdapter;

package/dist/adapters/claude-code.js

@@ -20,6 +20,12 @@ import { channelsFromConfig, shq, TokenUsageSchema } from "./types.js";
 // Phase 18's operator-price × tokens derivation, not a CLI claim.
 const MAX_SESSION_FILES = 20; // newest-first; a long-lived project dir can hold many sessions
 const MAX_SESSION_BYTES = 8_000_000; // per-file cap; a runaway JSONL cannot make the read unbounded
+// v1.75 T2 / OBS-137: current Claude Code workspace-trust prompt (2.1.218). The full question
+// distinguishes this startup gate from routine agent text; Enter accepts the selected trust option.
+export const CLAUDE_TRUST_DIALOG = {
+    fingerprint: "Quick safety check: Is this a project you created or one you trust?",
+    key: "Enter",
+};
 export function claudeSlug(real) {
     return real.replace(/[^A-Za-z0-9]/g, "-");
 }
@@ -54,13 +60,11 @@ export const claudeCode = {
     // and --mcp-config is VARIADIC — a positional after it is eaten as a config-file path, so another
     // flag must always follow the value, never the prompt.
     headlessCommand: (promptFile, model) => `claude -p "$(cat ${shq(promptFile)})" --model ${shq(model)} --permission-mode bypassPermissions --strict-mcp-config --mcp-config '{"mcpServers":{}}' --output-format text`,
-    // HYG-03: the residual first-entry dialog on an interactive TUI is the workspace TRUST dialog (not MCP
-    // config loading) — CLI-imposed, no flag to pre-accept, only store is claude's global last-writer-wins
-    // ~/.claude.json keyed on the exact path. Closed WON'T-FIX (decision B, 2026-07-10): tickmarkr writes nothing
-    // to that file (a seed races claude's own writes, nondeterministically). Amortizes to one operator dismissal
-    // per stable worktree path; blocked-pane paging surfaces it. Do NOT change this command to "fix" the dialog —
-    // see .planning/REQUIREMENTS.md HYG-03 and 21-02-LIVE-CHECK.md. Revisit if upstream ships a --trust flag.
+    // HYG-03 / OBS-137: the residual first-entry dialog is workspace trust, not MCP config loading.
+    // Claude's only store is global last-writer-wins ~/.claude.json, so tickmarkr still does not seed it;
+    // the daemon safely answers only the exact adapter-declared dialog once per slot.
     interactiveCommand: (promptFile, model) => `claude --model ${shq(model)} --strict-mcp-config --mcp-config '{"mcpServers":{}}' --permission-mode bypassPermissions "$(cat ${shq(promptFile)})"`,
+    trustDialog: CLAUDE_TRUST_DIALOG,
     resumeCommand: (sessionId, promptFile, model) => `claude -r ${shq(sessionId)} --model ${shq(model)} --strict-mcp-config --mcp-config '{"mcpServers":{}}' --permission-mode bypassPermissions "$(cat ${shq(promptFile)})"`,
     invoke(task, _cwd, a, ctx) {
         return { command: this.headlessCommand(ctx.promptFile, a.model) };

package/dist/adapters/codex.d.ts

@@ -1,8 +1,9 @@
-import { type TrustVerdict, type WorkerAdapter } from "./types.js";
+import { type TrustDialog, type TrustVerdict, type WorkerAdapter } from "./types.js";
 export declare function readCodexModelsCache(path?: string): {
     models: string[];
     fetchedAt?: string;
 };
+export declare const CODEX_TRUST_DIALOG: TrustDialog;
 export declare function seedCodexTrust(repoRoot: string, configPath?: string): TrustVerdict;
 export declare function hasCodexTrustedProject(text: string, root: string): boolean;
 export declare function codexConfigMcpServerNames(configPath?: string): string[];

package/dist/adapters/codex.js

@@ -89,6 +89,12 @@ const GITDIR_WRITABLE = `-c "sandbox_workspace_write.writable_roots=[\\"$(git re
 // -s/--sandbox workspace-write sandbox (deliberately NOT --dangerously-bypass-approvals-and-sandbox,
 // which would drop the sandbox). Listed by `codex --help` and `codex exec --help` (verified 2026-07-23).
 const CODEX_HOOK_TRUST = "--dangerously-bypass-hook-trust";
+// v1.75 T2 / OBS-137: current Codex workspace-trust prompt (0.144.6). The exact heading
+// is distinct from normal agent output; Enter accepts the selected "Yes, continue" option.
+export const CODEX_TRUST_DIALOG = {
+    fingerprint: "Do you trust the contents of this directory?",
+    key: "Enter",
+};
 // v1.22 T5 / OBS-16: codex keys trust on absolute path under [projects."<root>"] trust_level="trusted"
 // in ~/.codex/config.toml (CODEX_HOME relocates the dir). Worktrees inherit parent-project trust when
 // the REPO ROOT is trusted — seed the root once, cover every future worktree. Idempotent: a second
@@ -196,6 +202,7 @@ export const codex = {
     // v1.22 T5: seed [projects."<repoRoot>"] trust_level="trusted" so fresh worktrees never stall on
     // "Do you trust this directory?" (OBS-16). doctor-only side effect.
     trust: (repoRoot) => seedCodexTrust(repoRoot),
+    trustDialog: CODEX_TRUST_DIALOG,
     // v1.5 MODEL-01: file read only (no `codex models` subcommand exists, verified 2026-07-10).
     // Already fails OPEN to [] internally — advisory detection, unlike gates' fail-closed.
     listModels: async () => readCodexModelsCache().models,

package/dist/adapters/kimi.d.ts

@@ -20,6 +20,7 @@ export interface KimiInteractiveSeedResult {
     seedError?: string;
     sessionId?: string;
 }
+export declare const KIMI_INPUT_BOX: import("./types.js").InputBox;
 export declare function runKimiInteractiveSeed(opts: {
     driver: Pick<ExecutorDriver, "run" | "waitOutput" | "read">;
     slot: Slot;

package/dist/adapters/kimi.js

@@ -5,7 +5,7 @@ import { probeVersion } from "./claude-code.js";
 import { parseWorkerResult } from "./prompt.js";
 import { runInteractiveSeed } from "../run/interactive-seed.js";
 import { sh } from "../run/git.js";
-import { channelsFromConfig, MODEL_ID_RE, shq, TokenUsageSchema } from "./types.js";
+import { channelsFromConfig, declareInputBox, MODEL_ID_RE, shq, TokenUsageSchema } from "./types.js";
 // KIMI-03 → v1.58 T5: the "no harness-readable counter" block (research F-6, 2026-07-17) is
 // LIFTED for collectUsage — kimi 0.27.0 writes a wire journal per agent at
 // ~/.kimi-code/sessions/<wd>/session_<uuid>/agents/<agent>/wire.jsonl, and ~/.kimi-code/
@@ -118,6 +118,11 @@ export function confirmKimiSeedBanner(banner, assignedModel) {
     }
     return { ok: true, sessionId: kimiBannerSessionId(banner) };
 }
+// v1.75 T1 / OBS-136: this readiness line is rendered inside Kimi Code's bordered steady-state
+// input box. The adapter owns the fingerprint; the driver only consults declarations generically.
+export const KIMI_INPUT_BOX = declareInputBox("kimi", {
+    fingerprint: "Send /help for help information.",
+});
 // Shared launch-then-seed surface (T6) + banner confirm (T7/T2). One definition so the adapter
 // property and the daemon's generic runInteractiveSeed path cannot drift.
 const KIMI_SEED = {
@@ -165,6 +170,7 @@ export const kimi = {
     // prompt as one user turn. Banner model/session confirmation runs on the daemon's generic
     // runInteractiveSeed path via confirmBanner on KIMI_SEED (not a separate dispatch helper).
     interactiveSeed: KIMI_SEED,
+    inputBox: KIMI_INPUT_BOX,
     // v1.53 T3 resume — live-probed 2026-07-18: `-p` + `-S <id>` compose cleanly (no OBS-67-class
     // flag rejection) and the resumed session carries prior conversation state. `-S <id>` is the
     // deterministic form; `-c` rejected as primary — cwd-keyed, nondeterministic under worktree

package/dist/adapters/types.d.ts

@@ -82,6 +82,12 @@ export interface TrustDialog {
     key: string;
 }
 export declare function matchesTrustDialog(paneText: string, dialog: TrustDialog): boolean;
+export interface InputBox {
+    fingerprint: string;
+}
+export declare function declareInputBox(adapterId: string, inputBox: InputBox): InputBox;
+export declare function declaredInputBoxForWorkerName(workerName: string): InputBox | undefined;
+export declare function matchesInputBox(paneText: string, inputBox: InputBox): boolean;
 export interface WorkerAdapter {
     id: string;
     vendor: string;
@@ -105,6 +111,7 @@ export interface WorkerAdapter {
     contextUsage?(session: SessionRef): ContextUsage | null;
     trust?(repoRoot: string): TrustVerdict;
     trustDialog?: TrustDialog;
+    inputBox?: InputBox;
     hardcodedFlags?: {
         binary: string;
         flags: string[];

package/dist/adapters/types.js

@@ -30,6 +30,18 @@ export function modelAuthed(health, model, allowUnverifiedModels = false) {
 export function matchesTrustDialog(paneText, dialog) {
     return paneText.includes(dialog.fingerprint);
 }
+const inputBoxes = new Map();
+export function declareInputBox(adapterId, inputBox) {
+    inputBoxes.set(adapterId, inputBox);
+    return inputBox;
+}
+export function declaredInputBoxForWorkerName(workerName) {
+    const adapterId = /^.+-worker-(.+)-a\d+-.+$/.exec(workerName)?.[1];
+    return adapterId === undefined ? undefined : inputBoxes.get(adapterId);
+}
+export function matchesInputBox(paneText, inputBox) {
+    return paneText.includes(inputBox.fingerprint);
+}
 export function channelsFromConfig(adapterId, cfg) {
     const e = cfg.tiers[adapterId];
     if (!e)

package/dist/cli/commands/fleet.js

@@ -57,6 +57,14 @@ function withModeLine(yaml, mode) {
         return yaml.replace(/^routing:$/m, `routing:\n  mode: ${mode}`);
     return `routing:\n  mode: ${mode}\n${yaml}`;
 }
+function formatFleetSteering(cfg) {
+    const blocks = [];
+    if (cfg.review.prefer?.length)
+        blocks.push(`review:\n  prefer: ${JSON.stringify(cfg.review.prefer)}`);
+    if (cfg.consult.prefer?.length)
+        blocks.push(`consult:\n  prefer: ${JSON.stringify(cfg.consult.prefer)}`);
+    return blocks.length ? `${blocks.join("\n")}\n` : "";
+}
 // v1.51 T4: one gloss per routing mode on the fleet mode screen — mirrors the preset compiler.
 const MODE_GLOSS = {
     "partner-led": "every shape frontier · explore off",
@@ -83,7 +91,9 @@ export async function fleet(argv, cwd = process.cwd(), adapters = allAdapters(),
         const rm = resolveRunMode(cwd, { globalDir });
         const body = formatFleetPrint(cwd, { globalDir });
         const nl = body.indexOf("\n");
-        return `${body.slice(0, nl)}\n# mode: ${rm.mode.mode} (${rm.source})${body.slice(nl)}`;
+        // Steering comes from the same resolved config snapshot the editor consumes below,
+        // not from another parse of either raw overlay.
+        return `${body.slice(0, nl)}\n# mode: ${rm.mode.mode} (${rm.source})${body.slice(nl)}${formatFleetSteering(rm.cfg)}`;
     }
     if (!interactive)
         return { out: NON_TTY_MSG, code: 1 };

package/dist/drivers/herdr.d.ts

@@ -14,6 +14,7 @@ export declare class HerdrDriver implements ExecutorDriver {
     private deliverySerial;
     private dispatchLeases;
     private deliveredPanes;
+    private inputBoxes;
     private ws;
     private callerPane;
     private watches;

package/dist/drivers/herdr.js

@@ -1,4 +1,4 @@
-import { shq } from "../adapters/types.js";
+import { declaredInputBoxForWorkerName, matchesInputBox, shq } from "../adapters/types.js";
 import { PANE_IDENTITY_ENV, paneIdentityLine } from "../brand.js";
 import { createWorktree, sh } from "../run/git.js";
 import { herdrSealShellPrefix } from "./subprocess.js";
@@ -32,6 +32,7 @@ export class HerdrDriver {
     deliverySerial = Promise.resolve();
     dispatchLeases = new WeakMap();
     deliveredPanes = new WeakMap();
+    inputBoxes = new WeakMap();
     // VIS-10: the run's workspace id, captured once at construction (the daemon inherits it from the
     // operator's env before the driver is built). Required at slot() time, never in the constructor —
     // pickDriver and its unit test construct HerdrDriver without env, so slot() is the trust gate.
@@ -142,6 +143,7 @@ export class HerdrDriver {
         }
     }
     async slot(cwd, name, opts) {
+        const inputBox = declaredInputBoxForWorkerName(name);
         // T1 ownership contract: `opts.owned` (T2 call sites) names the pane canonically —
         // tickmarkr:<role>:<taskId>:<attempt>:<runId>. Without it, `name` passes through byte-identical
         // (today's legacy daemon/gates/consult shapes) — canonicalizeLegacyName (types.ts) is what lets
@@ -154,12 +156,13 @@ export class HerdrDriver {
         // Production dispatch names are canonical even when the gate call site supplies the already-
         // formatted name rather than SlotOpts.owned. Hold one lease across slot() → run(); legacy/manual
         // slots retain their existing allocation-only semantics for compatibility.
-        if (parseOwnedName(resolved))
-            return this.reserveDispatch(allocate);
+        const slot = parseOwnedName(resolved) ? await this.reserveDispatch(allocate) : await allocate();
+        if (inputBox)
+            this.inputBoxes.set(slot, inputBox);
         // group wins if both are set (a group tab is already stage-labeled; passing both is a caller bug).
         // label (without group) → dedicated labeled tab via tabSlot's third param: no groups-map entry, no
         // refcount, no groupSerial, no degrade latch — dedicated tabs have no shared state to guard (SUP-01).
-        return allocate(); // label undefined → defaults to name (today's behavior)
+        return slot; // label undefined → defaults to name (today's behavior)
     }
     // today's per-slot tab path, plus the VIS-04 orphan reap
     // label defaults to the slot name; group tabs pass the STAGE name instead — a first-member label
@@ -397,16 +400,19 @@ export class HerdrDriver {
                 // line only after two consecutive pane reads agree; an already-stable frame returns on the
                 // first fresh read without a timer. A changing pane is bounded and preserves OBS-85's
                 // fail-closed error instead of guessing from an adapter fingerprint.
-                const settled = await this.settleDeliveryLine(pane, slot.cwd, transcript);
+                const settled = await this.settleDeliveryLine(pane, slot.cwd, transcript, this.inputBoxes.get(slot));
                 transcript = settled.transcript;
                 if (!settled.ok) {
                     throw new Error(`herdr delivery clear failed — refusing to retype onto a corrupted line (OBS-85); pane transcript:\n${transcript}`);
                 }
-                // Clear the corrupted input line before retyping; a failed clear must NOT be retyped onto —
-                // corrupt-prefix + clean-retype would concatenate and false-verify by containment.
-                const cleared = await this.herdr(`pane send-keys ${shq(pane)} C-u`, slot.cwd);
-                if (cleared.code !== 0)
-                    throw new Error(`herdr delivery clear failed — refusing to retype onto a corrupted line (OBS-85); pane transcript:\n${transcript}`);
+                if (!settled.recognizedInputBox) {
+                    // Clear the corrupted shell input line before retyping; a failed clear must NOT be retyped
+                    // onto — corrupt-prefix + clean-retype would concatenate and false-verify by containment.
+                    // A stable adapter-declared input box is already an empty legitimate delivery target.
+                    const cleared = await this.herdr(`pane send-keys ${shq(pane)} C-u`, slot.cwd);
+                    if (cleared.code !== 0)
+                        throw new Error(`herdr delivery clear failed — refusing to retype onto a corrupted line (OBS-85); pane transcript:\n${transcript}`);
+                }
             }
             const typed = await this.herdr(`pane send-text ${shq(pane)} ${shq(cmd)}`, slot.cwd);
             if (typed.code !== 0)
@@ -423,20 +429,25 @@ export class HerdrDriver {
         }
         throw new Error(`herdr delivery corrupted after ${DELIVERY_ATTEMPTS} attempts — enter never pressed (OBS-85); pane transcript:\n${transcript}`);
     }
-    async settleDeliveryLine(pane, cwd, initialTranscript) {
+    async settleDeliveryLine(pane, cwd, initialTranscript, inputBox) {
         let transcript = initialTranscript;
         for (let readAttempt = 0; readAttempt < DELIVERY_SETTLE_READ_ATTEMPTS; readAttempt++) {
             const read = await this.herdr(`pane read ${shq(pane)} --source recent-unwrapped --lines ${DELIVERY_READ_LINES}`, cwd);
             if (read.code !== 0)
-                return { ok: false, transcript: read.stdout || transcript };
-            if (read.stdout === transcript)
-                return { ok: true, transcript };
+                return { ok: false, transcript: read.stdout || transcript, recognizedInputBox: false };
+            if (read.stdout === transcript) {
+                return {
+                    ok: true,
+                    transcript,
+                    recognizedInputBox: inputBox !== undefined && matchesInputBox(transcript, inputBox),
+                };
+            }
             transcript = read.stdout;
             if (readAttempt < DELIVERY_SETTLE_READ_ATTEMPTS - 1) {
                 await new Promise((resolve) => setTimeout(resolve, DELIVERY_SETTLE_POLL_MS));
             }
         }
-        return { ok: false, transcript };
+        return { ok: false, transcript, recognizedInputBox: false };
     }
     async deliveryReadMatches(pane, cmd, cwd) {
         const read = await this.herdr(`pane read ${shq(pane)} --source recent-unwrapped --lines ${DELIVERY_READ_LINES}`, cwd);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tickmarkr",
-  "version": "1.74.0",
+  "version": "1.75.0",
   "description": "Spec in, verified work out.",
   "type": "module",
   "license": "MIT",

package/skills/tickmarkr-auto/SKILL.md

@@ -15,8 +15,9 @@ When working in a multi-agent terminal environment, decide your role before star
 - **Orchestrator:** your session was started to execute the mission. Rename your own tab/pane `ORCH · <version>` (short labels: ≤20 chars, `ROLE · token`) and run the loop below.
 - **Supervisor with a live orchestrator:** do not start a second run. Relay the mission to the existing orchestrator with a [verified handoff](#verified-handoffs-agent-to-agent-messaging), then supervise it as OVERSEER.
 - **Primary session without an orchestrator:** rename your own tab `OVERSEER · <version>` and your agent `overseer`, spawn one child orchestration session with your host's launch form, label its tab `ORCH · <version>` and name its agent, give it the mission and these rules verbatim, then supervise it. Do not drive a duplicate single-tier run yourself. Before spawning, confirm any PREVIOUS orchestrator has stood down (monitors stopped, input box empty — dim ghost-text suggestions are UI, not queued input; ANSI-verify before alarming) and close its tab — the journal, records, and ledger hold the story; scrollback is disposable.
-  - **Claude Code:** `herdr agent start orchestrator --cwd <repo> --no-focus -- claude --permission-mode bypassPermissions`
-  - **Codex:** `herdr agent start orchestrator --cwd <repo> --no-focus -- codex --dangerously-bypass-approvals-and-sandbox` — the unsandboxed flag is REQUIRED, not optional: codex's `workspace-write` sandbox keeps `.git` refs read-only, so a sandboxed orchestrator's `tickmarkr run` dies at integration-branch creation (`git worktree add` cannot lock the ref). Do not downgrade this flag; the herdr pane and repo scope are the containment.
+  - **Spawning on current herdr is two-step** — the one-shot `agent start --cwd/--tab/--no-focus` form was removed in the herdr CLI redesign and now fails with `unknown option` (OBS-138). First create the pane: `herdr tab create --workspace <ws> --cwd <repo> --label "ORCH · <version>"` (tab create does not steal focus unless `--focus` is passed; parse `result.root_pane.pane_id` from its JSON), then start the agent in it:
+  - **Claude Code:** `herdr agent start orchestrator --kind claude --pane <root-pane-id> -- --permission-mode bypassPermissions`
+  - **Codex:** `herdr agent start orchestrator --kind codex --pane <root-pane-id> -- --dangerously-bypass-approvals-and-sandbox` — the unsandboxed flag is REQUIRED, not optional: codex's `workspace-write` sandbox keeps `.git` refs read-only, so a sandboxed orchestrator's `tickmarkr run` dies at integration-branch creation (`git worktree add` cannot lock the ref). Do not downgrade this flag; the herdr pane and repo scope are the containment.
   - **Auxiliary agents you spawn (consultants, reviewers, scouts) follow the same forms.** Never launch a claude session in plan mode or default permission mode for autonomous work — both stall on per-command approval prompts nobody is watching; claude is always `--permission-mode bypassPermissions` (tickmarkr's own adapter uses exactly this for workers, judges, and consults). A read-only codex consultant may use `--sandbox read-only`; any codex session that must touch git needs the unsandboxed flag above.
 
 Outside a multi-agent terminal environment, run the loop directly.

package/skills/tickmarkr-loop/SKILL.md

@@ -14,8 +14,9 @@ When working in a multi-agent terminal environment, decide your role before star
 - **Orchestrator:** your session was started to execute the mission. Rename your own tab/pane `ORCH · <version>` (short labels: ≤20 chars, `ROLE · token`) and run the loop below.
 - **Supervisor with a live orchestrator:** do not start a second run. Relay the mission to the existing orchestrator with a [verified handoff](#verified-handoffs-agent-to-agent-messaging), then supervise it as OVERSEER.
 - **Primary session without an orchestrator:** rename your own tab `OVERSEER · <version>` and your agent `overseer`, spawn one child orchestration session with your host's launch form, label its tab `ORCH · <version>` and name its agent, give it the mission and these rules verbatim, then supervise it. Do not drive a duplicate single-tier run yourself. Before spawning, confirm any PREVIOUS orchestrator has [stood down](#stand-down-mission-end-and-retirement) and close its tab.
-  - **Claude Code:** `herdr agent start orchestrator --cwd <repo> --no-focus -- claude --permission-mode bypassPermissions`
-  - **Codex:** `herdr agent start orchestrator --cwd <repo> --no-focus -- codex --dangerously-bypass-approvals-and-sandbox` — the unsandboxed flag is REQUIRED, not optional: codex's `workspace-write` sandbox keeps `.git` refs read-only, so a sandboxed orchestrator's `tickmarkr run` dies at integration-branch creation (`git worktree add` cannot lock the ref). Do not downgrade this flag; the herdr pane and repo scope are the containment.
+  - **Spawning on current herdr is two-step** — the one-shot `agent start --cwd/--tab/--no-focus` form was removed in the herdr CLI redesign and now fails with `unknown option` (OBS-138). First create the pane: `herdr tab create --workspace <ws> --cwd <repo> --label "ORCH · <version>"` (tab create does not steal focus unless `--focus` is passed; parse `result.root_pane.pane_id` from its JSON), then start the agent in it:
+  - **Claude Code:** `herdr agent start orchestrator --kind claude --pane <root-pane-id> -- --permission-mode bypassPermissions`
+  - **Codex:** `herdr agent start orchestrator --kind codex --pane <root-pane-id> -- --dangerously-bypass-approvals-and-sandbox` — the unsandboxed flag is REQUIRED, not optional: codex's `workspace-write` sandbox keeps `.git` refs read-only, so a sandboxed orchestrator's `tickmarkr run` dies at integration-branch creation (`git worktree add` cannot lock the ref). Do not downgrade this flag; the herdr pane and repo scope are the containment.
   - **Auxiliary agents you spawn (consultants, reviewers, scouts) follow the same forms.** Never launch a claude session in plan mode or default permission mode for autonomous work — both stall on per-command approval prompts nobody is watching; claude is always `--permission-mode bypassPermissions` (tickmarkr's own adapter uses exactly this for workers, judges, and consults). A read-only codex consultant may use `--sandbox read-only`; any codex session that must touch git needs the unsandboxed flag above.
 
 Outside a multi-agent terminal environment, run the loop directly.

package/skills/tickmarkr-overseer/SKILL.md

@@ -29,7 +29,7 @@ Requires `HERDR_ENV=1`; if unset, say so and stop.
    main name plus at most ONE hot-state token. Vocabulary: ORCH carries the milestone and progress
    fraction (`ORCH · v1.19 4/5`, updated on every task-done); WORKERS carries the task token (tickmarkr
    updates it). Never long context strings or ✓-chains.
-2. **Orchestrator**: Launch the orchestrator with your agent host. For Claude Code, use `herdr agent start orchestrator --cwd <repo> --no-focus -- claude --permission-mode bypassPermissions` (pin a strong model with `--model <m>` if the operator has a policy). For Codex, use `herdr agent start orchestrator --cwd <repo> --no-focus -- codex --dangerously-bypass-approvals-and-sandbox` (add `--model <m>` to specify the model). The unsandboxed flag is REQUIRED: codex's `workspace-write` sandbox keeps `.git` refs read-only, so a sandboxed orchestrator's `tickmarkr run` dies at integration-branch creation — do not downgrade it. Workers you never spawn — tickmarkr spawns its own visible worker panes. Auxiliary agents you do spawn (consultants, reviewers, scouts) follow the same forms: never launch a claude session in plan mode or default permission mode for autonomous work — both stall on per-command approval prompts nobody is watching; claude is always `--permission-mode bypassPermissions`, and a read-only codex consultant may use `--sandbox read-only`.
+2. **Orchestrator**: Launch the orchestrator with your agent host. Spawning on current herdr is two-step — the one-shot `agent start --cwd` form was removed in the herdr CLI redesign and now fails with `unknown option` (OBS-138): first create the pane with `herdr tab create --workspace <ws> --cwd <repo> --label "ORCH · <version>"` and parse `result.root_pane.pane_id` from its JSON, then start the agent in it. For Claude Code, use `herdr agent start orchestrator --kind claude --pane <root-pane-id> -- --permission-mode bypassPermissions` (append `--model <m>` after the `--` if the operator has a policy). For Codex, use `herdr agent start orchestrator --kind codex --pane <root-pane-id> -- --dangerously-bypass-approvals-and-sandbox` (add `--model <m>` to specify the model). The unsandboxed flag is REQUIRED: codex's `workspace-write` sandbox keeps `.git` refs read-only, so a sandboxed orchestrator's `tickmarkr run` dies at integration-branch creation — do not downgrade it. Workers you never spawn — tickmarkr spawns its own visible worker panes. Auxiliary agents you do spawn (consultants, reviewers, scouts) follow the same forms: never launch a claude session in plan mode or default permission mode for autonomous work — both stall on per-command approval prompts nobody is watching; claude is always `--permission-mode bypassPermissions`, and a read-only codex consultant may use `--sandbox read-only`.
 3. **Standing instructions travel as a brief FILE, never as pane text** — PTY input truncates at ~1024B and a
    truncated brief silently drops policy. Write the full brief to `<repo>/.tickmarkr/overseer/ORCH-BRIEF.md`
    (inside the tickmarkr state dir — already self-gitignored, no exclude step needed), then send one line: