@openrig/cli 0.5.13 → 0.5.15

package/daemon/assets/plugins/openrig-core/skills/applying-a-permission-policy/SKILL.md

@@ -1,117 +1,168 @@
 ---
 name: applying-a-permission-policy
-description: "Use when a permission policy is attached to a rig/seat (rig.yaml permission_policy:), or a user asks you to apply one, and you must translate that harness-neutral policy into the target harness's LIVE permission config — Claude settings.json / Codex config.toml / Pi run flags. Agent-driven and interactive: you read the policy, ground yourself in YOUR harness version, show the diff, and ask before writing. NOT for defining policies (that's the policy spec) and NOT a set-and-forget updater."
+description: >-
+  Use when a user asks an agent to configure OpenRig command permissions, reduce
+  repeated native approval prompts, or apply a selected rig/seat permission policy.
 metadata:
   openrig:
     stage: established
-    tested_against:                # VERSION STAMP — the whole point; re-verify at every harness bump
-      claude_code: "2.1.220 — VM-CONFIRMED 2026-08-03 (all cells)"
-      codex: "0.120.0 — VM-CONFIRMED 2026-08-03 (sandbox_mode/approval_policy; on-failure deprecated-but-accepted)"
-      pi: "0.83.0 — VM-CONFIRMED 2026-08-03: --approve/--no-approve govern PROJECT-RESOURCE TRUST, NOT tool permissions; Pi has NO permission surface (see Pi leg)"
-    verification_status: "Historical compatibility checks on the versions and dates in tested_against. These observations do not establish behavior on a later harness; verify its current format before applying the mapping."
+    docs_checked: "2026-09-25"
+    verification_status: "Verify the installed harness version and effective settings; a rule-parser result is not a native permission test."
 ---
 
 # Applying a permission policy
 
-## Translation is best-effort — read this first
-This skill provides the policy schema and OpenRig's best current understanding of how to translate it to each harness. That guidance is **best-effort and version-stamped**: harness permission surfaces change frequently across versions, so a mapping that is correct today can drift on a later release. When you apply a config-surface policy:
-- **Verify against the current docs for your installed harness version.** Treat this skill's tables as version-stamped starting points (see `tested_against`), not settled truth.
-- **Prefer testing in a protected or disposable environment first.** An incorrect write can lock a seat out or over-permit it, so confirm the format before applying it to a live seat.
-- **Hand-editing via the harness's own settings tools is always a valid path.** If a translation is uncertain, setting the policy directly — or asking the user to — is a legitimate outcome, not a failure.
-
-The reliable, deterministic parts are the **blunt instruments**: the YOLO full-bypass flag and the floor (Claude `acceptEdits` / Codex `workspace-write`), which ride the stable launch-flag surface. **Fine-grained config-rule translation — the allow/ask/deny prefix rules and the Codex posture — is the best-effort part** and carries no guarantee of exact fidelity. If a translation looks wrong, fall back to a blunt instrument or a direct hand-edit.
-
-## ⚠ Read this preamble first — why this is a SKILL, not a script (do not "improve" it into a deterministic updater)
-Harness permission formats are a **moving target**. The exact fact this skill is stamped against — Claude Code 2.1.220 permission rules are **prefix-only** (no flag-precise Bash match) — is *version-specific*; the rule grammar can change on the next point release. A hardcoded/deterministic updater would foot-gun the instant it does, and a wrong permission write can **lock a seat out or silently over-permit it**. So the maintained artifact is this caveated, version-stamped skill that an agent reads and applies **interactively, grounding in the live format** — never a code path that writes blindly. If you are tempted to replace this with a deterministic projector: that temptation is the bug this preamble exists to stop.
-
-## When to reach for this
-- A rig/seat has a policy attached (`rig.yaml … permission_policy: <file>`), and you're at **setup/preflight** — apply it (**mandatory for autonomous seats**, see the freeze warning).
-- A user asks "apply the Standard/Open/Locked/Operator policy here" or "set up permissions like <host>".
-- NOT for authoring policies (that's the policy spec) and NOT for daemon-level enforcement (OpenRig is agnostic — it injects nothing).
-
-## What you're translating (the boundary)
-INPUT = a **harness-neutral policy spec** (semantic action schema): `default_posture`, `floor`, `allow`/`ask`/`deny` as **semantic actions** (`push_to_remote`, `delete_files`, …), `destructive_class`, `source: builtin|custom`. You TRANSLATE that intent → the harness's live config. You never invent policy; you never pre-bake harness config into the spec.
-
-## The apply flow (interactive — the interactive-ask IS the product)
-1. **Read the policy spec** + its `source:` marker (built-in read-only from the package = copy-to-customize before editing; custom = user space).
-2. **PREFLIGHT (mandatory) — ground yourself in THIS environment:**
-   - **daemon-HOME must equal seat-HOME**, or a user-level Claude write never reaches the seat. Check; if they differ, **stop and ask** — do not write into the wrong HOME.
-   - Read the harness's **actual current config file** + confirm its schema against your installed version (`claude --version`, the live settings.json shape, codex/pi versions). If the format differs from this skill's `tested_against`, trust the LIVE format and flag the drift.
-3. **Derive the per-harness config** from the semantic actions (see the three legs below). Compute the Codex posture-collapse + Pi bit.
-4. **Show the concrete DIFF before writing** (old → new, per file). Never a blind write.
-5. **Resolve best-effort actions by the built-in's rule FIRST, then ask only on genuine ambiguity.** Best-effort destructive actions (`force_push`, `delete_files`, `delete_everything`, `read_secrets`) default to **`ask`** (never a `deny` that can't be enforced) — apply that default rather than stopping to question-back. Surface a question only when the policy is genuinely ambiguous or the seat is interactive. **For a same-prefix collision (`push_to_remote` vs `force_push`, both `Bash(git push:*)`), NEVER ask — apply the deterministic PREFIX-COLLISION rule in the Claude leg.** On a Claude-only fleet, if the policy denies `network_egress`, **say so explicitly**: it is NOT enforceable — Claude has no network gate, a `Bash(curl:*)`/`Bash(wget:*)` deny is best-effort, and `python`/`node` fetch bypasses it (live-verified: `urllib` reached HTTPS 200 under a curl/wget deny). Route real egress-gating to the Codex sandbox, or record it as advisory on Claude.
-6. **Back up the current config**, then write. Prefer the **user-level** file (`~/.claude/settings.json`) for durable policy — it's the clobber-resistant tier (deny wins across the union). **ALWAYS include `defaultMode: "acceptEdits"` in the Claude settings.json you write** — the floor key must be PRESENT in the file, or the config policy strips the floor.
-7. **Verify** the write landed and the seat can see it (re-read; confirm `defaultMode: "acceptEdits"` is present; for an autonomous seat, confirm no `ask` will freeze it).
-
-## Two surfaces — and this skill writes only ONE of them
-Permissions live on two surfaces with opposite stability — which is the whole reason for the agent-driven split:
-- **LAUNCH-FLAG surface — STABLE (>1yr) → OpenRig sets it DETERMINISTICALLY, NOT this skill.** The flags OpenRig passes at boot: Claude `--permission-mode` / `--dangerously-skip-permissions`; Codex sandbox/bypass flags; Pi `--approve`/`--no-approve`. **Two things live here, both OpenRig-owned: the FLOOR (minimum, by default) and YOLO (full bypass, opt-in).**
-- **CONFIG-FILE surface — CHAOTIC → NEVER deterministic → THIS SKILL'S DOMAIN.** Claude `settings.json` allow/ask/deny + defaultMode; Codex `config.toml` approval rules. This is where the Locked / Standard / Open policies get applied — agent-driven, version-stamped, caveated. (Pi has **no permission surface at all** — its `--approve`/`--no-approve` flags govern project-resource trust, not tool permissions (VM-confirmed Pi 0.83.0); a permission policy does not translate to Pi. See the Pi leg.)
-
-**THE FLOOR (launch-flag, OpenRig-set — this skill KNOWS it, never WRITES it):** Claude `--permission-mode acceptEdits` (keep it); **Codex workspace-only** (stop forcing `danger-full-access` — ≈ Codex's own default ≈ setting-nothing = agnostic); Pi `--no-approve` (a project-resource-trust floor — Pi has no permission gate; see the Pi leg). **One consistent unconditional minimum**. A chosen config-file policy layers ON TOP of the floor.
-
-**YOLO MODE (launch-flag, OpenRig-set, opt-in — for people done with permissions):** OpenRig boots every seat with the full-bypass flag (Claude `--dangerously-skip-permissions`; Codex full-bypass). **When YOLO is on, CONFIG-FILE POLICY IS MOOT** — this skill does NOT apply a config policy (or explicitly notes it's overridden by the flag); don't fight the bypass. YOLO is OpenRig's deterministic job, not this skill's. (**Operator = YOLO mode** — subsumed.)
-
-## The three harness legs (skill-owned mapping knowledge, version-stamped + caveated)
-
-### Claude — `~/.claude/settings.json` (user-level, clobber-resistant)
-- **Action → prefix map** (VM-CONFIRMED @ Claude 2.1.220, 2026-08-03 — every row exercised through real Claude tool calls). `fidelity`: `clean` = the prefix honors intent; `best_effort` = prefix-only can't fully enforce → prefer `ask`, never a `deny` that lies.
-
-  | Semantic action | `permissions.*` prefix rule(s) | fidelity |
-  |---|---|---|
-  | push_to_remote | `Bash(git push:*)` | clean |
-  | force_push | ask the flag-first forms `Bash(git push --force:*)` `Bash(git push -f:*)` `Bash(git push --force-with-lease:*)`; base `Bash(git push:*)` follows push_to_remote (see the collision rule) | **best_effort** — catches flag-FIRST force; flag-LAST (`git push origin main --force`) leaks |
-  | create_pr | `Bash(gh pr create:*)` | clean |
-  | publish_package | `Bash(npm publish:*)` `Bash(pnpm publish:*)` `Bash(yarn publish:*)` `Bash(cargo publish:*)` | clean (multi-pattern) |
-  | merge_or_release | `Bash(git merge:*)` `Bash(git tag:*)` `Bash(gh release:*)` | best_effort — protected-branch semantics not expressible |
-  | delete_files | `Bash(rm:*)` | **best_effort** — target-first leak (`rm <t> -rf` slips); scope not boundable → default `ask` |
-  | delete_everything | `Bash(rm -rf:*)` `Bash(rm -fr:*)` | **best_effort** — same leak → default to `ask`, never a silent deny |
-  | reset_or_discard_vcs | `Bash(git reset:*)` `Bash(git clean:*)` `Bash(git checkout:*)` `Bash(git branch -D:*)` | best_effort |
-  | drop_persistent_store | `Bash(dropdb:*)` `Bash(docker volume rm:*)` (+ project-specific) | best_effort (open-ended) → `ask` |
-  | rig_up / rig_down | `Bash(rig up:*)` / `Bash(rig down:*)` | clean — ⚠ an `ask` here FREEZES an autonomous seat |
-  | mutate_topology | `Bash(rig add:*)` `Bash(rig remove:*)` `Bash(rig rename:*)` | clean |
-  | run_toolchain | `Bash(npm:*)` `Bash(npx:*)` `Bash(pnpm:*)` `Bash(yarn:*)` `Bash(node:*)` `Bash(tsc:*)` `Bash(vitest:*)` `Bash(jest:*)` | clean (the known set) |
-  | run_arbitrary_shell | `Bash(*)` — this IS the `default_posture` knob | clean |
-  | network_egress | `Bash(curl:*)` `Bash(wget:*)` (partial) | **best_effort** — Claude has NO native network gate; real floor = Codex sandbox (route it there) |
-  | read_secrets | `Read(./.env)` `Read(~/.ssh/**)` `Read(**/*secret*)` (path-scoped) | **best_effort** — path-based, not secret-aware |
-  | install_dependencies | `Bash(npm install:*)` `Bash(pip install:*)` `Bash(brew install:*)` | clean |
-- **PREFIX-ONLY caveat** — cannot flag-precisely gate (`force_push` vs `push_to_remote` both match `Bash(git push:*)`); **target-first leak** (`rm <target> -rf` slips a `Bash(rm -rf*)` deny). Actions marked `best_effort` in the taxonomy (`force_push`, `delete_files`, `read_secrets`) → surface the caveat and prefer `ask` over a `deny` you can't enforce (a `deny` that leaks lies).
-- **PREFIX-COLLISION resolution (deterministic — resolve it, NEVER ask).** When two actions collapse to the SAME base prefix with different dispositions — the canonical case is `push_to_remote`=allow + `force_push`=ask, both basing to `Bash(git push:*)` — you cannot put that base prefix in both `allow` and `ask`. Resolve without a question-back:
-  - **Push allowed + force asked (Standard / Open):** put the base `Bash(git push:*)` per push's disposition (`allow`), AND add the flag-first force forms to `ask` (`Bash(git push --force:*)`, `Bash(git push -f:*)`, `Bash(git push --force-with-lease:*)`). RECORD in your diff that this is best-effort — flag-LAST force (`git push origin main --force`) leaks past the flag-first ask rules and runs under the allow. Don't block on it.
-  - **Push denied (Locked):** `Bash(git push:*)` in `deny` already covers force — no conflict.
-  This is a FIXED rule, not a user decision — apply it and note the leak.
-- **`defaultMode: "acceptEdits"` is the FLOOR and MUST be written into every Claude settings.json you produce.** It is not a knob you tune, but it IS a key you must always include — a config policy that omits it strips the floor. Live-confirmed schema: `{"permissions":{"allow":[],"ask":[],"deny":[],"defaultMode":"acceptEdits"}}`.
-
-### Codex — `~/.codex/config.toml` (via the existing `codex_config_fragment` splice path — reuse, don't invent)
-- Codex is **per-posture, not per-action**: the whole action set **collapses** to the nearest `{sandbox_mode, approval_policy}` pair. **Built-in → posture** (VM-CONFIRMED @ Codex 0.120.0, 2026-08-03 — each posture applied to a real seat: read-only blocked writes+egress, workspace-write allowed writes/blocked egress, danger-full-access allowed both):
-
-  | Built-in | `sandbox_mode` | `approval_policy` | Collapse note (surface this to the user) |
-  |---|---|---|---|
-  | Locked | `read-only` | `on-request` | read-only blocks all writes + egress, prompts on escalation — OVER-restricts `edit_files` vs the Claude floor |
-  | Standard ⭐ | `workspace-write` | `on-request` | workspace writes allowed, egress blocked by sandbox, prompt on escalation — UNDER-gates `create_pr` (no per-action ask; folds into the prompt) |
-  | Open | `danger-full-access` | `on-failure` | everything allowed, prompt only on failure — OVER-permits the `destructive_class` vs the Claude ask-gates |
-  | YOLO | *(flag-surface — full-bypass launch flag, NOT config.toml)* | — | not a config write; the rip-out adapter owns it |
-
-  Valid values (VM-confirmed @ Codex 0.120.0): `sandbox_mode` ∈ {read-only, workspace-write, danger-full-access}; `approval_policy` ∈ {on-request, on-failure, never, untrusted, granular} (`granular` observed live at 0.120.0). ⚠ `on-failure` (used by Open) is marked **deprecated** in 0.120.0 help but still accepted — revisit if a future Codex drops it.
-- Surface the collapse: "Codex can't gate per-action; this policy becomes sandbox=X approval=Y, which over-permits Z / under-permits W."
-
-### Pi — NO permission surface at all (VM-confirmed Pi 0.83.0 — the earlier "approval bit" reading was WRONG)
-- **Pi has no tool-permission gate — neither a config store NOR a permission-posture flag.** Live verification confirmed: Pi 0.83.0's `--approve`/`-a` and `--no-approve`/`-na` govern **project-resource trust** (whether project extensions/resources appear in RPC `get_commands`), **NOT** tool-call approval. There is no sandbox or permission mechanism to map a policy onto.
-- So a permission policy **does not translate to Pi**. Do NOT say `--approve` = "auto-approve all", or that a posture "collapses to a Pi approval bit" — that was an earlier misread, corrected here. This skill computes **nothing** for Pi.
-- Whatever the OpenRig adapter sets (`--no-approve` by default) is a project-resource-trust floor, not a permission posture — not this skill's to derive.
-- Behavior to know (resource-trust, NOT permission-gating): under default ask / `--no-approve`, a project resource is silently **absent** from RPC `get_commands`; it appears under `--approve`. Never mistake that for permission enforcement.
-- **Say this to the user for a Pi seat:** "Pi has no permission-policy surface today — the policy applies to Claude/Codex; Pi runs under its project-resource-trust flag unchanged."
-
-## ⚠ The two warnings that save real fleets
-- **`ask` FREEZES an autonomous/headless Claude seat** — it hangs on the first non-edit op. So for autonomous fleets apply **Open/YOLO** (or a policy with no asks) **at setup** — the floor only covers edits; bash still asks without an applied policy. Never leave an autonomous seat on a policy with live `ask` gates. (Pi has no permission ask at all — see the Pi leg; Codex prompts per its posture, not per-action.)
-- **daemon-HOME ≠ seat-HOME** silently voids user-level writes — always the preflight check above.
-
-## The built-ins — DEFINED by the shipped policy spec files (translate FROM the spec; POINT, don't COPY)
-The four built-in policies ship as read-only spec files (`source: builtin`, `policy_schema_version: 1`) — they are the AUTHORITATIVE definitions. **Do NOT restate their action-sets here.** The selected spec supplies the action sets; a second copy in this skill can diverge. Read the attached/selected spec and translate its semantic-action sets (`default_posture` / `allow` / `ask` / `deny` / `destructive_class`) per the three legs above.
-- **Locked / Standard ⭐ / Open** — `surface: config`. Translate to the Claude / Codex config surface. **Standard ⭐** is the recommended default: routine dev incl. push allowed, the outward/release/history-rewriting acts (`create_pr`, `publish_package`, `merge_or_release`, `force_push`) ASK, destructive_class ASKs. Standard has live `ask` gates → not for a headless-autonomous Claude seat (use Open or YOLO). Locked = deny-default whitelist; Open = allow-default, destructive_class asks only.
-- **YOLO** — `surface: flag`, `launch_posture: full_bypass`. This skill does **NOT** translate it — it points at OpenRig's deterministic flag-surface opt-in (the rip-out YOLO setting). When YOLO is on, any config-surface policy is MOOT (the bypass overrides it). (Subsumes the former "Operator".)
-
-Built-ins ship read-only in the package (canonical names); copy-to-customize into user space (custom, user-named); the `source:` frontmatter marker travels on copy. Onboarding presents these as a required MENU (explain each + ask which; no-choice = the floor).
-
-## Boundary (keeps this the only maintained piece)
-This skill TRANSLATES + APPLIES only. It does **not** define policies (the policy spec does) and does **not** enforce at the daemon (OpenRig is agnostic). Spec = *what* (neutral intent); this skill = *how, on this harness version* (the caveated, version-stamped translation).
+Configure the user's chosen permissions in the harness that runs the agent.
+OpenRig operating posture, native command rules, sandbox access and launch flags
+are separate controls. A command rule grants execution capability, not authority
+to invent tasks, publish work or change another host.
+
+## Choose the intended scope
+
+Use an existing explicit choice; otherwise explain these options and ask which
+the user wants. Do not ask again for routine steps already authorized.
+
+| Choice | What the agent configures |
+| --- | --- |
+| Keep prompts | Preserve current native settings and handle requests when they arise. |
+| Remember selected commands | Add native allow rules for the chosen family or narrower verbs, leaving other rules and sandbox settings intact. |
+| Broader permissive operation | Explain filesystem/network exposure and configure only the explicitly selected native mode and compatible launch settings. |
+
+**Allowing the whole `rig` family covers all its verbs**, including lifecycle,
+topology/config changes and commands that can launch other processes. It is not
+a read-only grant. Offer narrower prefixes such as `rig ps` or `rig queue list`
+when that better fits the request. Do not widen a choice to arbitrary shell
+execution, an entire interpreter or a generic shell wrapper.
+
+## Apply the choice for the user
+
+1. Identify the target seat, executable/version, launch settings and config roots:
+   `HOME`, `CODEX_HOME` or `CLAUDE_CONFIG_DIR` as applicable. They may differ between
+   operator, daemon and seat. Resolve the intended user's/project's files before
+   writing; do not repair another home to make the paths agree.
+2. Read relevant existing permission rules and managed restrictions. Select the
+   intended scope: one project or the user's sessions. Consult installed native
+   help and official references below when formats differ.
+3. Prepare a concrete diff. Preserve deny/ask rules, approval/sandbox posture,
+   hooks, auth, MCP, model settings and unrelated values. An allow must not erase
+   a stricter rule or managed requirement. Report a real conflict instead of
+   silently bypassing it.
+4. Back up touched files and merge only authorized additions, avoiding duplicates.
+   The agent performs these edits; hand-editing is an option, not a required user
+   chore. Apply the selected scope without another conversational permission round.
+   Native enforcement still applies.
+5. Read back the diff and validate the format. Confirm how that version loads
+   changes. If it requires a new session, preserve work and use its supported
+   resume path within the user's authority; a file write does not prove that an
+   existing conversation loaded it.
+6. Verify an ordinary matching operation twice in the target conversation and
+   check that an unrelated command gained no matching rule. Use harmless reads,
+   not destructive probes. Report effective settings and remaining prompts.
+   A parser match alone does not establish native behavior.
+
+For rollback, remove only this setup's additions, preserving later unrelated
+edits. Builtin policy specs remain read-only; customize in user space.
+
+## Codex command rules
+
+Codex command rules can allow matching commands outside the sandbox without
+another prompt. They leave other sandbox/network settings unchanged.
+Check `codex --version` and `codex execpolicy check --help`.
+See the [official rules reference](https://learn.chatgpt.com/docs/agent-configuration/rules).
+
+Choose the destination **before writing**, according to the user's scope:
+
+- **This project only:** verify the installed version supports project rules and
+  derive this session's actual project/worktree config root and trust state.
+  Current docs describe `<repo>/.codex/rules/` in a trusted project config layer.
+  Use that layer only after confirming it is supported, active and trusted for
+  the intended project. If any of those facts is unsupported or unverified,
+  report the limitation and leave user-layer rules unchanged; do not silently
+  mark a project trusted or substitute a user-wide allowance.
+- **Explicitly user-wide:** use `rules/` under the target user's actual
+  `CODEX_HOME` (normally `~/.codex/rules`). This can affect other projects using
+  that home. The TUI's remember-allow action also writes a user-layer rule;
+  do not use it to implement a project-only request.
+
+Merge the chosen rule into a `.rules` file in the selected layer. The prefix
+itself has no project restriction; even a project-layer rule does not constrain
+which targets an allowed `rig` command can affect:
+
+```python
+prefix_rule(pattern = ["rig"], decision = "allow")
+```
+
+For narrower access use `["rig", "ps"]` or `["rig", "queue", "list"]`, adjusting
+examples. For absolute-path invocations, derive the target seat's actual `rig`
+executable and add that exact path as a separate prefix; a bare rule does not
+match it. Never copy another machine's path.
+
+```sh
+codex execpolicy check --pretty --rules /absolute/path/to/openrig.rules -- rig ps --json
+codex execpolicy check --pretty --rules /absolute/path/to/openrig.rules -- printf permission-check
+```
+
+Inspect matches, not only exit status; repeat `--rules` for other effective
+files. A matching `prompt` or `forbidden` overrides allow. Verify rule loading
+in the target conversation after the required reload. For project-only scope,
+confirm the layer is active there and absent from an unrelated project's active
+layers. An evaluator given an explicit `--rules` file proves matching, not that
+scope or automatic loading. Existing user-wide rules may already permit the
+same command: preserve and disclose them, attribute matches, and do not claim
+project isolation or remove those rules without a separate authorized choice.
+
+The standalone evaluator checks supplied argv; native shell parsing can split
+ordinary commands/chains first. A raw `zsh -lc` non-match does not prove that
+native `rig && rig` fails. Verify both surfaces without broadening the rule to
+`bash`, `sh`, `node` or generic wrappers.
+
+## Claude Code command rules
+
+Check `claude --version` and [official permission syntax](https://code.claude.com/docs/en/permissions).
+Merge the chosen entry into existing `permissions.allow`; this fragment is not
+a replacement settings file:
+
+```json
+{
+  "permissions": {
+    "allow": ["Bash(rig *)"]
+  }
+}
+```
+
+Current syntax uses ` *` for a command family; `Bash(rig:*)` is also supported.
+Narrower examples are `Bash(rig ps *)` and `Bash(rig queue list *)`.
+Preserve `deny`/`ask` entries and `defaultMode`; do not add `Bash(*)` or switch
+to bypass to resolve a mismatch. Inspect native `/permissions` and verify the
+actual command spelling.
+
+Choose scope using the [settings reference](https://code.claude.com/docs/en/settings):
+`.claude/settings.local.json` for personal project settings,
+`.claude/settings.json` for deliberately shared project settings, or
+`settings.json` under the target's `CLAUDE_CONFIG_DIR` (normally `~/.claude`).
+Confirm the effective project root, especially for worktrees. Keep personal
+settings out of commits. Managed restrictions and sandbox/network controls
+still apply; a Bash allow is not a general network policy.
+
+## Existing policies and broader modes
+
+For a named policy, read its actual `permission_policy` spec and `source` marker.
+Translate intended actions using supported native controls; do not collapse all
+Codex policies to a single posture or claim shell patterns perfectly express
+semantic actions such as force-push. Preserve stricter rules. If exact translation
+is unavailable, explain the remaining choice instead of selecting broader access.
+
+For explicitly chosen broader operation, inspect
+`rig policy current --spec <user-owned-rig.yaml>` and the compatible getting-started guide's **Opt-in permissive
+operation** section. OpenRig's Codex `builtin:yolo` supplies `-s danger-full-access`;
+it does not itself select `approval_policy` and replaces a named
+`codex_config_profile` argument. Claude's corresponding launch flag is
+`--dangerously-skip-permissions`. Neither a resource `profile: default` nor OpenRig
+operating posture is a native permission policy. Do not change shipped defaults
+or assume editing a launch spec changes an existing seat.
+
+A headless seat may wait at a native prompt. Arrange an answer path or choose
+suitable command rules; unattended work is not implicit consent to bypass.
+For Pi, the previously supported `--approve`/`--no-approve` surface concerns
+project-resource trust, not shell permissions; verify its installed capabilities
+instead of treating those flags as a command allowlist.

package/daemon/assets/plugins/openrig-core/skills/messaging-the-human/SKILL.md

@@ -42,16 +42,44 @@ Check readiness: configured, enabled, active, ready, reason, and next action.
 `indeterminate` is not ready. Follow the reported next inspection; do not enable
 or reconfigure a connector merely to make the check pass.
 
-Write the decision, evidence, and continuation in a body file. The sole outbound
-human-message primitive is:
+Author for the person reading on a phone: a short subject in `--summary`, then
+one complete brief in `--body-file`. State why it matters, your recommendation
+and material tradeoff, the bounded action if approved, and the choice requested.
+For an update, state the user-visible outcome and “No action needed.” Aim for
+roughly 100–150 words; this is guidance, not a semantic validator. Keep technical
+continuation, exact candidate/revision and evidence on the owning agent row and
+in the durable artifact behind `--evidence-ref`. A local path is not a phone link
+and Markdown evidence files are not automatically attached.
+
+For example, a synthetic brief could say:
+
+> The repaired status view is ready. I recommend updating this instance; live
+> status will briefly pause. Sessions will be preserved. Approve this instance
+> update, or hold? Supporting test detail follows in this thread.
+
+This example grants no authority. Choose `--human-intent decision` for a request
+or `--human-intent update` for a quiet FYI. Omission retains legacy decision
+behavior; words such as “FYI” and tags do not change intent.
+
+The sole outbound human-message primitive is:
 
 ```bash
 rig queue create --destination <entityId>@external \
-  --summary "<decision or update>" --body-file <packet-file> \
+  --human-intent decision --summary "<short subject>" --body-file <brief-file> \
   --evidence-ref <durable-evidence> --verify --json
 ```
 
-If an existing agent-owned row must wait, block it on the **new live qitem ID**
+An optional `--human-detail-file <path>` supplies one coherent supplemental
+reply in the same thread. Announce its purpose in the brief; the product also
+marks that a detail reply follows. The primary must already contain the complete
+scope, options and action. Do not split an agent dump blindly or move the essential
+choice into overflow. Rendering checks every part and its accessibility fallback
+before posting; an oversized request is refused with a field-specific correction,
+never silently clipped. Shorten the brief or related detail as directed. Inspect
+the failed row, then deliberately cancel/replace the authored request if its
+content needs correction; a timeout alone is never a reason to replace it.
+
+If an existing agent-owned row must wait for a **decision**, block it on the **new live qitem ID**
 (`rig queue block <work-id> --on <human-qitem-id> ...`), not on the human address.
 Completion of the human qitem resumes its dependants. Blocking on the human as
 well would issue another notification for the same decision.
@@ -66,7 +94,14 @@ or blindly resend because verification timed out.
 rig queue transitions <qitem-id>
 ```
 
-A correlated reply binds to that exact human and qitem and records the resolution
+For `update`, confirmed complete delivery may close the delivery obligation.
+It creates no approval obligation and cannot be used as a decision blocker.
+Delivered updates remain queryable for Feed; a failure or ambiguous send stays
+separate. A root message alone does not prove supplemental delivery. Retries
+reconcile stable part identities and send only missing parts. An FYI reply is
+not a human decision.
+
+For a decision, a correlated reply binds to that exact human and qitem and records the resolution
 that resumes the owner. Check the recorded result before claiming the decision
 arrived; a delivery receipt alone is not acceptance.
 

package/daemon/assets/plugins/openrig-core/skills/openrig-skills/SKILL.md

@@ -71,6 +71,7 @@ These are auto-delivered to every rig; their name+description are already in you
 ### Load when your role or task calls for it (repo-shipped, profile-selected)
 These ship in the OpenRig repo and reach a seat when its profile selects them. To use one, select it in your profile's `uses.skills`, or open it directly at `packages/daemon/specs/agents/shared/skills/core/<skill>/SKILL.md`.
 
+- **openrig-software-factory** — a user wants continuing reviewed work: manual/team, queue-only orchestration or an explicit Workflow; add one or two seats to the running starter when useful, with context, ownership, concurrency, spend and permission limits. Load `rig context get skills/core/openrig-software-factory/SKILL.md`.
 - **openrig-architect** — authoring a rig or topology (NOT for changing OpenRig itself — that's `openrig-builder`).
 - **openrig-cmux** — driving the `cmux` terminal provider.
 - **openrig-herdr** — opening/managing seat terminals via the default proof-gated provider.

package/daemon/context-packs/onboarding-width/manifest.yaml

@@ -1,6 +1,6 @@
 # Public capability and source-map width for the default onboarding walk.
 name: onboarding-width
-version: "0.5.13"
+version: "0.5.15"
 taxonomy: world
 purpose: "The public capability map and source map referenced by the default onboarding walk."
 files:

package/daemon/context-packs/onboarding-width/public-what-you-can-do.md

@@ -11,7 +11,7 @@ not know exists.** So the point of what follows is not skill. It is recognition:
 lands, something rhymes, and you go check instead of building.
 
 Eighty-one-plus top-level verbs ship (capability canon refreshed through
-`capability-delta-v0.5.13`). This marker describes the pack's teaching, not publication or live
+`capability-delta-v0.5.14-r6`). This marker describes the pack's teaching, not publication or live
 adoption. Read this once for shape, and let it make you suspicious that a thing already exists.
 Model-divergence proclamations are live product (trust them; pins use canonical
 model IDs).
@@ -100,8 +100,10 @@ unseen — which makes messages the one delivery channel that never gets skipped
   someone lists the stream before starting.
 - **`rig terminal open <view>`** — bring every live agent in a rig, mission or slice up as real
   typeable tiles at once.
-  In the TUI, **TERMINALS** lists Saved and Derived views. Select a view to inspect its members,
-  layout and pages; preview is passive. **Open in Herdr** deliberately opens the inspected plan
+  In the TUI, **TERMINALS** keeps Saved views prominent and Derived groups collapsed until
+  expanded. Names load before detailed readiness. Select a view to inspect its members,
+  layout and pages; preview is passive. Saved membership requires deliberate setup.
+  **Open in Herdr** deliberately opens the inspected plan
   and reports opened, absent or degraded members. Help or a side trip returns to the same preview.
 - **`rig walk <seat> --through <files> --pace <n>`** — deliver context pieces through
   file-backed terminal paste. When the current native generation record resolves, the complete
@@ -295,18 +297,29 @@ because someone lost work once.
 - **`rig doctor`** — is the *installation* wired up correctly, or are you chasing a bug that is
   really a broken install. **`rig preflight`** asks whether this machine can run OpenRig at all.
 - **Bare `rig`** — the same TUI for first setup, daemon-down startup and ordinary work.
+  A positive observation of a running rig enters ordinary work without a repeated Startup choice.
+  With no running rig, the chooser remains; slow or unverified detection does not mean stopped.
+  A late observation never takes navigation away after you have chosen another route.
   **?** Help, **w** Skip and **L** Local reading remain usable during slow probes or loading,
   and when the daemon is down or unverified. Local reading shows selected disk intent with
   provenance and read errors; it does not substitute for live queue, execution or topology data.
   These reading controls do not start a daemon or resume seats; skipping an already accepted
   operation does not cancel it.
-  Enter starts only the selected daemon; then choose rigs and individual seats, with kernel
-  recommended first. Previously occupied seats default to their authoritative conversation.
+  In startup, Enter starts only the selected daemon; then choose rigs and individual seats,
+  with kernel recommended first. Previously occupied seats default to their authoritative conversation.
   Missing history, ambiguous identity and unavailable authentication have distinct explanations;
   a fresh conversation requires a separate named confirmation. Declining launches nothing.
   **o** opens the existing native terminal here to handle its prompts; detach to return.
   If a fresh start paused before context delivery, **c** finishes that same occupant’s context.
   **r** refreshes actual state and **d** expands diagnostics. **S** returns to startup from work.
+  Ordinary views keep the navigator usable through page/rig changes and slow or failed reads.
+  Topology offers rig selection first, then loads that rig before unrelated fleet details.
+  While a new page loads, useful prior content can remain under its original scope label with
+  no action targets. First visits retain navigation and local loading/error feedback.
+  During refresh or failure, previously loaded content in the same scope stays useful, with its last
+  successful read time and Retry. A failed Feed source retains its contribution while healthy
+  sources update; confirmed denial, empty results or removals clear affected content. Late responses cannot
+  replace another page or project's data, and surviving selection and scroll stay in place.
 - **`rig crash-cart`** — the read-only daemon verdict and saved-state discovery used by that UI.
   An unavailable runtime, unauthorized endpoint or unreadable state is not an empty instance.
 - **`rig snapshot`** — take a restore point **before** the risky thing. `snapshot list` shows what
@@ -362,10 +375,23 @@ scheme, and nothing downstream can see it.
   `rig tui`, **PROJECTS** → choose a project → mission → slice shows the work story with row
   drill-in, current source and Escape back. Project IDs and roots distinguish equal display
   names; an unavailable project read does not substitute another project's work.
-- **TUI ATTENTION** — inspect **Action required** separately from **Updates**. Requests and
-  explicit human blockers show what decision is needed and what work it unblocks; outcome and
-  health updates retain their source and history. Open detail and evidence, then Back. Reading
-  does not approve or deliver a request, and queue closure alone does not accept an outcome.
+  The mission overview puts outcomes, current work/owner, blockers, next dependencies and
+  readable slice boxes before process prose. Accepted required proof judgments determine
+  completion; attached evidence, quiet activity and handoff do not. Reopened and assigned
+  unfinished work stays open even when no unassigned next slice exists. Actual custody and
+  planned ownership remain distinct, as do outcome completion and mission lifecycle or
+  publication. Unknown owners and eligibility stay unknown. Narrow views keep boxes and scroll.
+  Malformed mission or slice sources stay visible as local unavailable entries while healthy
+  neighbors remain readable. Open the affected source to inspect it, then refresh after correction.
+- **TUI Feed** — inspect **Human requests** separately from **Updates**, with the scope shown as
+  **Instance / All humans**. Explorer lists these categories; their entries appear in the
+  content pane. Human destinations and explicit human blockers show the recipient,
+  relevant project, needed decision and work it unblocks; agent priority alone is not a human
+  request. Confirmed delivered quiet FYIs remain in a bounded update window after their delivery
+  obligation closes, with a receipt and **No action needed**. Failed or ambiguous delivery is not
+  delivered history. Existing outcome and health updates retain their source and history.
+  Open detail and evidence, then Back. Reading does not approve or deliver a request, and queue
+  closure alone does not accept an outcome. Direct `attention`, `needs` and `feed` commands remain.
 - **`rig scope slice ls` / `show` / `create`** — the same at the altitude where work is actually
   buildable. `show` gives you intent, frontmatter and children without guessing which of five files
   to open.
@@ -400,6 +426,7 @@ scheme, and nothing downstream can see it.
   policy acceptance, higher outcome judgment and publication remain separate decisions.
 - **`rig workspace doctor` / `validate`** — does the daemon agree with you about where the work
   tree is, and which files are missing the frontmatter their kind requires.
+- **OpenRig Software Factory** — when a first team needs continuing reviewed work or one or two more seats in its running rig; covers queue-only and optional Workflow paths, context/work ownership, wake and token costs, permissions, and optional custom-rig authoring. Discover the bundled recipe with `rig context show skills/core/openrig-software-factory --json`, then load `rig context get skills/core/openrig-software-factory/SKILL.md`; use guidance compatible with the installed build.
 - **`rig context work-install --project … --mission … --slice … [--deliver]`** — resolves the ordered
   System World, topology, and Project World plan. Add `--runtime` to see the composed managed skill
   loadout, `--apply-skills` to reconcile its owned harness projection, or `--deliver` to emit the
@@ -492,6 +519,9 @@ reaching them is ordinary work rather than an escalation.
   pending, failed, or indeterminate delivery leaves the row intact. Inspect the same row's
   transitions before retrying. `rig send` remains agent-terminal delivery only.
 - **When project policy calls for human judgment or an update, load `messaging-the-human`.**
+  Use `rig context get skills/core/messaging-the-human/SKILL.md` for a complete decision brief,
+  related supplemental detail and explicit quiet-update intent. An FYI creates no approval
+  obligation; writing “FYI” in ordinary request prose does not select that intent.
   System World teaches these mechanics; Project World decides when and why they apply.
   Existing `<entityId>@host` blockers are internal custody labels mapped through the registry
   to the external participant, not a separate outbound route.
@@ -552,22 +582,26 @@ your circumstances is configuration, and the ones that are not, another agent ca
   Open the instance row for one continuous cross-rig agent table with pod separators and material
   `RECENT` transitions; drill into a rig, mission, slice, or agent without losing the owning
   identity. Use the mission's workflow/packet view and Specs purpose/source to understand work.
-  In **Specs**, selection previews purpose, contents and provenance; Enter opens details.
+  In **Specs**, kind groups start collapsed and expand explicitly. Selection previews purpose,
+  contents and provenance; Enter opens details.
   **View current source** reads disk within the explicitly configured readable roots. Relative
   Markdown links resolve against that source; headings open with a labelled starting point.
   Escape or `back` restores the caller's selection and scroll. Detail uses the full width on
   narrow terminals. The TUI command `read <root>/<path>#heading` also opens a named source;
   missing, denied, binary, truncated and missing-heading results stay explicit. HTTP(S) links
   show a selectable destination without opening a browser; `v` enables terminal text selection.
-  Tab completes commands and snapshot arguments; Recent opens original events. **CONFIG**
-  browses instance/work roots, context, display, waiting, recovery, activity and advanced settings;
+  Tab completes commands and snapshot arguments; Recent opens original events. **System** opens
+  to instance **Health**, with **Configuration** and **Connections** beneath it. Contextual Health
+  remains available in the owning rig or agent view. **Configuration** browses instance/work
+  roots, context, display, waiting, recovery, activity and advanced settings;
   Slack and people are one category. Select a category, use right/left to change panes,
   Enter for full value/default/source/scope, `/` to search labels and keys, and Escape or `back`
   to return. Long detail scrolls; `v` enables terminal text selection. `refresh` reads the current
   view again without changing settings or verifying delivery. Sources & coverage names exclusions,
   daemon/CLI identity and the client timezone separately. Resolved settings do not prove runtime
-  adoption. The legacy `connections` command still inspects gateway services and human routes;
-  dated verification does not prove delivery. `timezone` gives persistent local-time guidance.
+  adoption. Direct `config` and `connections` commands remain; Connections inspects gateway
+  services and human routes. Dated verification does not prove delivery. `timezone` gives
+  persistent local-time guidance.
   **?** opens full-screen Help with command grammar and examples; Escape returns to the caller.
   **`rig tui commands`** lists everything it can do without launching it. **`rig ui open`** is
   unmaintained, best-effort, and replaced by the TUI, so never diagnose product behaviour from

package/daemon/context-packs/skills/agent-operated-software/manifest.yaml

@@ -1,7 +1,7 @@
 # GENERATED by scripts/generate-context-packs.mjs — DO NOT EDIT.
 # Package-time projection of a canonical skill; regenerated every build.
 name: "agent-operated-software"
-version: "0.5.13"
+version: "0.5.15"
 taxonomy: "skills"
 purpose: "Use when designing, building, operating, or diagnosing an ongoing application whose live backend or control loop includes OpenRig agents, including applications with a Markdown, YAML, or JSON agent control plane or a thin surface over specialist agent roles."
 files:

package/daemon/context-packs/skills/agent-operated-workflows/manifest.yaml

@@ -1,7 +1,7 @@
 # GENERATED by scripts/generate-context-packs.mjs — DO NOT EDIT.
 # Package-time projection of a canonical skill; regenerated every build.
 name: "agent-operated-workflows"
-version: "0.5.13"
+version: "0.5.15"
 taxonomy: "skills"
 purpose: "Use when a bounded real-world procedure has a deterministic happy path but brownfield, variable, or partially knowable state; when an operation must resume from verified evidence; or when deciding whether agent judgment or ordinary code should own a procedure's control loop."
 files: