arcane-cli 0.29.2 → 0.29.3

package/dist/assets/.arcane/governance/agent-approved-paths.md

@@ -36,9 +36,9 @@ Define one block per agent. Each agent gets its own isolated workspace root, plu
 
 **Conventions:**
 
-- Each agent operates from its own `workspace-{agent}` root — never a shared tree — so concurrent agents don't collide.
-- Grant the narrowest access that lets the agent do its job (research agents: `Read-only`; builders: `Read/Write`).
-- The operator's primary working tree is never an approved agent path.
+- **Each agent operates from its own `workspace-{agent}` root. Enforcement: explicitly advisory prose (ARC-023) — this describes a convention for the operator's own agent-fleet setup; no check in this repository verifies that any agent actually runs from an isolated workspace root.** — never a shared tree — so concurrent agents don't collide.
+- **Grant the narrowest access that lets the agent do its job. Enforcement: explicitly advisory prose (ARC-023) — which access level counts as "narrowest" for a given agent is a judgment call; no check verifies the grant against actual job scope.** (research agents: `Read-only`; builders: `Read/Write`).
+- **The operator's primary working tree is never an approved agent path. Enforcement: explicitly advisory prose (ARC-023) — no check cross-references this file's approved-path rows against the operator's actual working-tree path to catch a violation.**
 
 ---
 
@@ -46,15 +46,15 @@ Define one block per agent. Each agent gets its own isolated workspace root, plu
 
 When a request targets a repo that is expected to be inside an approved root:
 
-1. Perform read-only discovery first (list/find/read).
-2. If path exists, continue task without asking for duplicate authorization.
-3. If not found, request the exact path from the operator.
-4. Do not search outside approved roots.
+1. **Perform read-only discovery first. Enforcement: explicitly advisory prose (ARC-023) — the same behavior universal-agent-rules.md rule 20 covers; no check verifies that an agent actually performs read-only discovery before proceeding.** (list/find/read).
+2. **If path exists, continue task without asking for duplicate authorization. Enforcement: explicitly advisory prose (ARC-023) — whether an agent treats prior discovery as sufficient and skips a redundant ask is a judgment call; no check verifies this occurred.**
+3. **If not found, request the exact path from the operator. Enforcement: explicitly advisory prose (ARC-023) — the same "before claiming no access" behavior universal-agent-rules.md rule 20 covers; no check verifies an agent actually asks rather than assuming access is absent.**
+4. **Do not search outside approved roots. Enforcement: explicitly advisory prose (ARC-023) — no check inspects an agent's actual filesystem access to confirm it stayed within approved roots.**
 
 ---
 
 ## Change Control
 
-- Add/remove paths only through documentation update in this file.
-- Reflect material scope changes in [[agents/agent-policies|agent-policies.md]].
-- If access boundaries meaningfully change risk posture, record an ADR.
+- **Add/remove paths only through documentation update in this file. Enforcement: explicitly advisory prose (ARC-023) — no check cross-references actual agent path grants against this document to catch an undocumented change.**
+- **Reflect material scope changes in [[agents/agent-policies|agent-policies.md]]. Enforcement: explicitly advisory prose (ARC-023) — whether a scope change is "material" enough to require an update, and whether the update happens, both depend on agent judgment; no check cross-validates this file against agent-policies.md.**
+- **If access boundaries meaningfully change risk posture, record an ADR. Enforcement: explicitly advisory prose (ARC-023) — `check:adr-references` verifies that ADR citations elsewhere resolve, not that a change to this file actually triggered a new ADR.**

package/dist/assets/.arcane/governance/agent-work-queue-model.md

@@ -18,6 +18,8 @@ Defines a single markdown-first queue for assigning work across businesses, whil
 - Queue rows include execution host and approval gate so autonomy stays explicit.
 - {BUSINESS_NAME} starts first; same structure extends to AcmeStore and other businesses.
 
+**This document specifies a queue design that has not been implemented anywhere in this codebase — no queue instance exists yet, and this document's own frontmatter is still `status: draft` (see the `## Next Action` section below). Enforcement: explicitly advisory prose (ARC-023) — every principle, field, and rollout step that follows is advisory guidance only, since no code, spell, or platform check exists yet to enforce any of it.**
+
 ---
 
 ## Queue Principles

package/dist/assets/.arcane/governance/authentication-strategy.md

@@ -24,16 +24,19 @@ This document provides rationale for authentication decisions across an agent ru
 - Bearer tokens (PATs, API keys) can be replayed if intercepted once
 - Prefer challenge/response authentication (SSH, OAuth) when available
 - When bearer tokens are required, store them as environment variables via a secret-reference pattern
+- **Enforcement: explicitly advisory prose (ARC-023) — these two practices govern a downstream operator's own authentication mechanism and secret storage; no code or spell in this repository observes which mechanism or pattern is actually used.**
 
 **2. Credential Isolation**
 - Each service/agent uses separate credentials with minimal required scope
 - Credentials stored outside version control (`.env`, system keychains, secret managers)
 - Rotation/revocation must be straightforward without config file changes
+- **Enforcement: explicitly advisory prose (ARC-023) — credential scoping, storage location, and rotation ease are properties of a downstream operator's own credential stores; no script in this repository inspects a consumer's `.env` files, keychains, or secret managers.**
 
 **3. Defense in Depth**
 - Authentication is one layer; combine with network isolation (loopback bind, VPN/tunnel)
 - Audit logging for all credential operations
 - Regular review of active credentials and permissions
+- **Enforcement: explicitly advisory prose (ARC-023) — network isolation, audit logging, and credential review are carried out entirely in the operator's own infrastructure; this repository ships no gateway, log pipeline, or scheduled check that observes them.**
 
 ## Authentication Contexts
 
@@ -47,6 +50,8 @@ This document provides rationale for authentication decisions across an agent ru
 - Private key remains local at `~/.ssh/id_*` (`600` permissions)
 - Git remote configured to use the SSH URL
 
+**Enforcement: explicitly advisory prose (ARC-023) — this decision and its mechanism and implementation-notes steps configure a downstream operator's own Git host account and local filesystem; Arcane ships no code that checks which credential type is registered with the Git host or that verifies key-file permissions.**
+
 **Why SSH Over PAT:**
 
 1. **Bearer Token Replay Risk**
@@ -95,6 +100,8 @@ This document provides rationale for authentication decisions across an agent ru
 - Store the key as an environment variable (e.g. in a `.env` file)
 - Reference it in config via a secret-reference rather than embedding the value
 
+**Enforcement: explicitly advisory prose (ARC-023) — this decision and its mechanism and implementation-notes steps configure a downstream operator's own model-provider account and local secret storage; Arcane ships no code that verifies key scope, storage location, or file permissions.**
+
 **Why Scoped Keys Over Personal Keys:**
 
 1. **Organizational Boundary**
@@ -132,6 +139,8 @@ This document provides rationale for authentication decisions across an agent ru
 - Reference it from config rather than embedding the literal value
 - Require it in the HTTP header: `Authorization: Bearer <token>`
 
+**Enforcement: explicitly advisory prose (ARC-023) — this decision and its mechanism and implementation-notes steps describe a gateway component operated outside this repository; Arcane's own codebase contains no gateway server to bind an address, validate a token, or enforce this header requirement.**
+
 **Why Token Auth:**
 
 1. **Stateless Authentication**
@@ -187,6 +196,8 @@ auth = { provider = "token", token = { source = "env", id = "SECRET_NAME" } }
 SECRET_NAME=actual-secret-value-here
 ```
 
+**Enforcement: explicitly advisory prose (ARC-023) — Arcane ships no config-schema validator or lint rule in this repository that checks a downstream repo's config for inline secrets versus this reference pattern; adoption depends on the operator's own discipline.**
+
 **Benefits:**
 - Config files are version-controlled without exposing secrets
 - Secrets are rotated by editing the secret store only (no config changes)
@@ -212,6 +223,8 @@ Regular review (quarterly):
 - [ ] Grep the codebase for hardcoded secrets (should find zero matches)
 - [ ] Review the security audit log for any anomalous credential operations
 
+**Enforcement: explicitly advisory prose (ARC-023) — this checklist runs on operator judgment and cadence; no scheduled task, spell gate, or CI check in this repository executes or verifies these items.**
+
 ---
 
 ## References

package/dist/assets/.arcane/governance/hardening-checklist.md

@@ -20,6 +20,8 @@ Step-by-step hardening for a Linux host that will run an autonomous agent runtim
 
 See also: [Threat Model](threat-model.md), [Authentication Strategy](authentication-strategy.md)
 
+**Enforcement: explicitly advisory prose (ARC-023) — every item below configures a downstream operator's own Linux host, a target outside anything Arcane's own CI or spells can inspect; completion is tracked only in the operator's own security audit log, not by any Arcane mechanism.**
+
 ---
 
 ## Pre-Checklist: Know Your Environment

package/dist/assets/.arcane/governance/naming-conventions.md

@@ -24,9 +24,9 @@ How machines, AI agents, and systems are named across your projects. This is a *
 
 Two rules govern every name in the system.
 
-**The Naming Test:** if an established industry term exists for the thing, use the real term — a universe name has to be *earned* by the absence of one. Two dialects result: magic grammar for the personas and vocabulary, plain functional payloads for everything technical. Corollary: the more autonomous the tool, the more boring its name should be.
+**The Naming Test:** if an established industry term exists for the thing, use the real term — a universe name has to be *earned* by the absence of one. Two dialects result: magic grammar for the personas and vocabulary, plain functional payloads for everything technical. Corollary: the more autonomous the tool, the more boring its name should be. **Enforcement: explicitly advisory prose (ARC-023) — whether a real industry term already exists is a judgment call; no check screens a candidate name against this test.**
 
-**The IP rule:** persona names must be **public domain or your own coinage**. No living people. No franchise-coined characters — a borrowed name is owned by whoever coined it, no matter how well it fits. Before any name ships, run the four checks: who coined it; does an estate still do business in it; do same-audience products already claim it (search live — don't trust memory); and what is its first association in each market you operate in, said aloud in every language your team speaks.
+**The IP rule:** persona names must be **public domain or your own coinage**. No living people. No franchise-coined characters — a borrowed name is owned by whoever coined it, no matter how well it fits. Before any name ships, run the four checks: who coined it; does an estate still do business in it; do same-audience products already claim it (search live — don't trust memory); and what is its first association in each market you operate in, said aloud in every language your team speaks. **Enforcement: explicitly advisory prose (ARC-023) — the four checks depend on human judgment and live research; `spell-scry` can perform this exact check but nothing requires it before an agent name is assigned, and the naming step in `src/modules/naming.ts` validates only that a typed name is non-empty.**
 
 Names should be memorable, distinctive, and never repeated across categories.
 
@@ -70,6 +70,7 @@ Agents get persona names paired with a role. The persona name is the primary ide
 - An agent may serve multiple projects but keeps the same name and role everywhere.
 - If an agent's role changes significantly, it gets a new name (don't reuse names for different roles).
 - If you collaborate with other operators who run their own Arcane instances, reserve names across instances so they don't collide.
+- **Enforcement: explicitly advisory prose (ARC-023) — no code checks persona-name uniqueness within a roster, across projects, or across Arcane instances; `agent-schema.ts`'s roster validation checks each entry's shape only.**
 
 ### Default Roster (Arcane personas)
 
@@ -107,9 +108,9 @@ The table above is the human-readable roster; agent definitions (`agents/*.yaml`
 
 **Role consolidation (ADR-031):** Framework-specific dev roles (.NET, Blazor, React, Flutter, MAUI) are consolidated into generic roles (Full-Stack, Frontend, Mobile). The framework/stack is a prompt parameter, not an agent identity. This scales without needing a new agent for every technology.
 
-**Naming policy:** All agent names should be positive characters — beings, not objects — with a clear thematic connection to their role.
+**Naming policy:** All agent names should be positive characters — beings, not objects — with a clear thematic connection to their role. **Enforcement: explicitly advisory prose (ARC-023) — character and tone are a judgment call; `agent-schema.ts` validates that a name is a non-empty string, not its thematic content.**
 
-**Canonical agent identity domain:** Use a single domain for all agent and tool Git identities — e.g., `merlin@{OPERATOR_DOMAIN}`. See [[DECISIONS#ADR-028|ADR-028]] for the full attribution model.
+**Canonical agent identity domain:** Use a single domain for all agent and tool Git identities — e.g., `merlin@{OPERATOR_DOMAIN}`. See [[DECISIONS#ADR-028|ADR-028]] for the full attribution model. **Enforcement: explicitly advisory prose (ARC-023) — no commit-trailer or identity check verifies that every agent/tool email shares one domain; `git-conventions.md`'s Required Commit Trailers table validates trailer presence, not the author-email domain.**
 
 ### Power Levels (ADR-032)
 

package/dist/assets/.arcane/governance/new-business-setup.md

@@ -38,6 +38,8 @@ Have the following ready:
 
 ## Phase 1 — Project & Repo Setup
 
+**Every step in this phase is performed by hand in your version-control host's UI. Enforcement: explicitly advisory prose (ARC-023) — project, repo, and template creation happen entirely on the external VCS host, which no script or CI check in this repo observes or verifies.**
+
 ### Step 1.1 — Create the Project
 
 1. Open your version-control host's project organization
@@ -82,6 +84,8 @@ In your docs/ops repo, copy `ventures/_template/` to `ventures/{venture}/` as a
 
 ## Phase 2 — Infrastructure & Security
 
+**The security checklist and docs-repo clone in this phase are unverified operator actions. Enforcement: explicitly advisory prose (ARC-023) — no script or CI check in this repo confirms these steps were completed or completed correctly.**
+
 ### Step 2.1 — Confirm Security Baseline
 
 Before running the agent runtime for any new business, verify:
@@ -101,6 +105,8 @@ git clone <docs-repo-url> ~/code/{repo}
 
 ## Phase 3 — Agent Runtime Configuration
 
+**Steps 3.1, 3.2, and 3.4 (runtime install, configuration, and the dry-run test) are manual operator actions. Enforcement: explicitly advisory prose (ARC-023) — no script or CI check in this repo verifies the agent runtime was installed, configured, or dry-run tested before live use.**
+
 ### Step 3.1 — Install the Agent Runtime (if not already installed)
 
 See [[agents/installation|Runtime Installation]].
@@ -113,6 +119,8 @@ See [[agents/installation|Runtime Installation]].
 
 ### Step 3.3 — Agent Git Identity Setup
 
+**Registering an agent's persona, Git identity, and `.mailmap` entry, and running the verify command below, are self-administered. Enforcement: explicitly advisory prose (ARC-023) — no script or CI check validates `.mailmap` entries or roster registration; the verify command is a manual smoke test, distinct from `spell-commit-work`'s Step 8 gate, which governs the commit action itself, not agent registration.**
+
 Every agent that will produce commits for this business must have a registered Git identity **before making its first commit**. See [[git-conventions#agent-attribution-model|Git Conventions — Agent Attribution]].
 
 For each agent assigned to this business:
@@ -141,6 +149,8 @@ Run the agent runtime in dry-run / observation mode first. Do not let it execute
 
 ## Phase 4 — Legal Baseline
 
+**Every step in this phase happens outside Arcane — entity structuring, tax registration, and ToS review. Enforcement: explicitly advisory prose (ARC-023) — these are legal and tax determinations on external systems that no script or CI check in this repo can observe or verify.**
+
 ### Step 4.1 — Business Entity
 
 - Confirm operations run under {LLC_NAME}
@@ -161,6 +171,8 @@ Run the agent runtime in dry-run / observation mode first. Do not let it execute
 
 ## Phase 5 — First Launch Checklist
 
+**This launch checklist is self-attested against external systems (the selling platform, the agent runtime's monitoring). Enforcement: explicitly advisory prose (ARC-023) — no script or CI check in this repo confirms any item below; the operator marks each complete by judgment.**
+
 - [ ] At least 10 product listings live (or equivalent for non-e-commerce)
 - [ ] Order/fulfillment workflow tested end-to-end
 - [ ] The agent runtime running and monitored (see `agents/monitoring.md`)

package/dist/assets/.arcane/governance/poc-management-pattern.md

@@ -34,7 +34,7 @@ ventures/        Prototypes      features/     business repo
 
 #### Concept (`ventures/<name>/`)
 
-The permanent record of every idea — good, bad, or deferred. Ideas are **never deleted**. Frontmatter `status` tracks where each idea is in its lifecycle.
+The permanent record of every idea — good, bad, or deferred. Ideas are **never deleted**. **Enforcement: explicitly advisory prose (ARC-023) — no code in this repo checks for or blocks deletion of files under `ventures/<name>/`; this depends on human/agent discipline.** Frontmatter `status` tracks where each idea is in its lifecycle.
 
 **Required files:**
 - `overview.md` — What is it? Why does it matter? Who is it for? (Minimum viable concept doc.)
@@ -45,7 +45,7 @@ The permanent record of every idea — good, bad, or deferred. Ideas are **never
 - `stories.json` — If architecture produced story breakdown
 - Screenshots in `assets/screenshots/YYYY-MM-DD/` (linked from docs)
 
-**What does NOT belong here:** Code files (`.js`, `.ts`, `.py`, `.html`, `.css`, etc.). The docs repo is a documentation-only repository.
+**What does NOT belong here:** Code files (`.js`, `.ts`, `.py`, `.html`, `.css`, etc.). The docs repo is a documentation-only repository. **Enforcement: explicitly advisory prose (ARC-023) — [[governance/agent-policies|Agent Policies]]'s repo-routing rule states the general version of this restriction; this file restates it for the docs repo specifically rather than asserting independent enforcement, and no lint/CI check here rejects code file extensions.**
 
 #### Prototype (Prototypes repo)
 
@@ -57,7 +57,7 @@ Small, self-contained, throwaway code that demonstrates a concept. Prototypes ar
 
 **Folder convention:** `<name>/` at repo root (e.g., `magic-prompt/`, `self-healing-poc/`)
 
-**No governance overhead:** No PRs required, no work items, no branch policies. This is a sandbox. Commit directly to main.
+**No governance overhead:** No PRs required, no work items, no branch policies. This is a sandbox. Commit directly to main. **Enforcement: explicitly advisory prose (ARC-023) — this is a deliberate low-ceremony policy choice, not a constraint with a pass/fail check to enforce.**
 
 #### Feature (`features/{id}-{slug}/`)
 
@@ -116,7 +116,7 @@ Ideas evolve. Track iterations without complex versioning:
 
 2. **Screenshots:** Save evidence of each meaningful state to `assets/screenshots/YYYY-MM-DD/` per [[assets/screenshots/README|screenshot conventions]]. Link from the changelog.
 
-3. **Git history is the version control.** Don't create `v1/`, `v2/` folders. The idea folder is a living document — git tracks every change.
+3. **Git history is the version control.** Don't create `v1/`, `v2/` folders. The idea folder is a living document — git tracks every change. **Enforcement: explicitly advisory prose (ARC-023) — no check inspects idea folders for `v1/`/`v2/`-style version directories; this depends on human/agent discipline.**
 
 ---
 
@@ -128,10 +128,10 @@ No formal gate. If someone (human or agent) wants to see the idea running, gener
 
 ### Concept/Prototype → Feature
 
-All of these must be true:
+All of these must be true: **Enforcement: explicitly advisory prose (ARC-023) — no spell checks these four criteria before a `status: promoted` transition; `spell-plan` satisfies only the fourth criterion (a PRD exists) and does not gate the others, and `spell-manifest`'s promotion flow uses a separate idea-book status marker, not this file's ventures frontmatter.**
 - [ ] A tracker work item exists (User Story or Feature)
 - [ ] A business track owns it (maps to a `ventures/<name>/` folder or a project)
-- [ ] A target code repo exists (or needs to be created; repo creation is human-only)
+- [ ] A target code repo exists (or needs to be created; repo creation is human-only) **Enforcement: explicitly advisory prose (ARC-023) — [[governance/agent-policies|Agent Policies]] is the canonical statement of this rule (Spectator-level, human-only); this restates it rather than asserting independent enforcement, and no code under `src/commands/` implements repo creation for a check to gate in the first place.**
 - [ ] The idea has at least a PRD (`spell-plan` output)
 
 When promoted: update the idea's frontmatter to `status: promoted` with `promoted_to` metadata. The idea folder stays in arcane forever as the historical record.
@@ -166,7 +166,7 @@ In the vibe coding era, a well-written spec can regenerate a prototype in minute
 |---|---|
 | [[governance/development-methodology\|Spell Loop]] | Ideas enter the loop at `spell-plan`. Concept stage is pre-loop exploration. |
 | Repo routing | Code → code repos. Docs → the docs repo. Prototypes → Prototypes repo. |
-| Repo governance | Repo creation is human-only. Prototypes repo is a single shared repo, not per-POC. |
+| Repo governance | Repo creation is human-only. **Enforcement: explicitly advisory prose (ARC-023) — [[governance/agent-policies\|Agent Policies]] is the canonical statement of this rule; this table restates it rather than asserting independent enforcement.** Prototypes repo is a single shared repo, not per-POC. |
 | Feature folders | Feature planning artifacts live in the target code repo under `features/`. |
 | `ventures/<name>/` | Business-level docs for shipped products. Ideas that graduate here are fully promoted. |
 

package/dist/assets/.arcane/governance/records-conventions.md

@@ -64,16 +64,16 @@ superseded_by: path/to/replacement.md
 
 Rules:
 
-- `superseded_by` is required. A tombstone that does not say what replaced it is
+- **`superseded_by` is required. Enforcement: explicitly advisory prose (ARC-023) — no script checks a superseded document's frontmatter for this field.** A tombstone that does not say what replaced it is
   worse than no tombstone: it tells the reader to stop without telling them
   where to go.
-- If a document was superseded by *several* documents, list all of them.
-- If a document was retired without replacement, say so explicitly
-  (`superseded_by: none — <reason>`), rather than leaving the field off.
-- Do not delete the original content. The tombstone is a header, not a
+- **If a document was superseded by *several* documents, list all of them. Enforcement: explicitly advisory prose (ARC-023) — no script cross-checks the list against the document's actual replacements.**
+- **If a document was retired without replacement, say so explicitly
+  (`superseded_by: none — <reason>`), rather than leaving the field off. Enforcement: explicitly advisory prose (ARC-023) — no script distinguishes this explicit form from an omitted field.**
+- **Do not delete the original content. Enforcement: explicitly advisory prose (ARC-023) — no diff check runs when a tombstone is added, so nothing confirms the body underneath is untouched.** The tombstone is a header, not a
   replacement for the body.
-- **Add the reverse link on the replacement**, so the relationship is
-  discoverable from either end. A reader who finds only the current document
+- **Add the reverse link on the replacement, so the relationship is
+  discoverable from either end. Enforcement: explicitly advisory prose (ARC-023) — no script checks that a `supersedes` field exists on the named replacement.** A reader who finds only the current document
   should still be able to see what it replaced:
 
   ```yaml
@@ -91,7 +91,7 @@ Supersession and deletion are different decisions and must not be conflated.
 Marking a document superseded is a routine editorial act. Deleting it is not.
 
 **Deletion is prohibited without an explicit, recorded decision when any of the
-following is true:**
+following is true. Enforcement: explicitly advisory prose (ARC-023) — none of the four conditions below is checked by a script; `content_sensitivity`'s value is schema-validated, but nothing gates deletion on it.**
 
 - The document records a decision, agreement, or obligation that other parties
   relied on.
@@ -101,8 +101,8 @@ following is true:**
 - The repository is marked `content_sensitivity: sensitive` and the document's
   own existence is part of the record.
 
-When deletion *is* approved, record it: what was deleted, when, on whose
-authority, and why. A deletion nobody can account for later is indistinguishable
+**When deletion *is* approved, record it: what was deleted, when, on whose
+authority, and why. Enforcement: explicitly advisory prose (ARC-023) — no script checks that a deletion record exists after the fact.** A deletion nobody can account for later is indistinguishable
 from data loss.
 
 **Arcane does not ship a retention schedule.** Retention periods are

package/dist/assets/.arcane/governance/threat-model.md

@@ -50,7 +50,7 @@ See also: [Hardening Checklist](hardening-checklist.md), [Authentication Strateg
 | Supply chain (package installs) | **Mitigated** | Global package installs use a user-owned directory, not a system directory. No `sudo` during install. |
 | Inter-machine communication | **Mitigated** | Encrypted mesh tunnel/VPN. No direct unencrypted LAN communication. |
 | Token/credential exposure (at rest) | **Mitigated** | Token stored as an env var in a `.env` file. Config uses a secret reference, not plaintext. State directory is mode `700`. |
-| Credential committed to version control | **Not mitigated** | Storage conventions are documented (secret references, `.gitignore` entries for `.env`/`*.pem`/`*.key`) but **no detection exists** — no scanner runs on the commit path, the push path, or in CI, and nothing verifies the conventions were followed. `spell-security-review`'s secrets step is an on-demand agent read, not a control. Treat every credential that reaches a commit as compromised and rotate it — history, existing clones, and remote backups all survive deleting the file (EF-35). |
+| Credential committed to version control | **Not mitigated** | Storage conventions are documented (secret references, `.gitignore` entries for `.env`/`*.pem`/`*.key`) but **no detection exists** — no scanner runs on the commit path, the push path, or in CI, and nothing verifies the conventions were followed. `spell-security-review`'s secrets step is an on-demand agent read, not a control. **Treat every credential that reaches a commit as compromised and rotate it — history, existing clones, and remote backups all survive deleting the file (EF-35). Enforcement: explicitly advisory prose (ARC-023) — rotating a leaked credential happens at the third-party provider that issued it; no Arcane mechanism can perform or verify that a specific credential was actually rotated.** |
 
 ---
 
@@ -94,11 +94,11 @@ This policy defines which collaborators can access which project data. Enforce b
 
 ### Enforcement
 
-- Project-level isolation is the primary boundary: each business is a separate project with its own access control.
-- A partner's access is limited to a single business's project, through a dedicated org only.
-- An operations co-owner receives access only to their business's project and relevant platform accounts — not to infrastructure configs.
-- System-level and multi-business reference documentation stays private to the owner unless explicitly shared.
-- New collaborator access requires an explicit entry in this table and an entry in the security audit log.
+- **Project-level isolation is the primary boundary: each business is a separate project with its own access control. Enforcement: explicitly advisory prose (ARC-023) — isolation depends on the owner actually configuring separate platform projects/repos with distinct access control; `spell doctor`'s `checkPlatformBranchPolicy` (src/modules/platform-policy.ts) verifies only branch merge-method policy and `checkDelegations` (src/commands/doctor.ts) verifies only agent power-level delegations, so no Arcane check queries collaborator or team membership to verify a given business's project is actually isolated.**
+- **A partner's access is limited to a single business's project, through a dedicated org only. Enforcement: explicitly advisory prose (ARC-023) — depends on the owner correctly scoping the partner's platform account to a single org; no Arcane check inspects collaborator or team membership to verify a partner's actual access is so limited.**
+- **An operations co-owner receives access only to their business's project and relevant platform accounts — not to infrastructure configs. Enforcement: explicitly advisory prose (ARC-023) — depends on the owner correctly scoping platform-account grants; no Arcane check inspects collaborator permissions to verify infrastructure configs are actually excluded.**
+- **System-level and multi-business reference documentation stays private to the owner unless explicitly shared. Enforcement: explicitly advisory prose (ARC-023) — depends on the owner's manual sharing decisions; no Arcane check inspects document visibility or sharing state to verify this boundary holds.**
+- **New collaborator access requires an explicit entry in this table and an entry in the security audit log. Enforcement: explicitly advisory prose (ARC-023) — a manual documentation discipline; no Arcane check verifies this table was updated or that a corresponding security audit log entry exists, and the security audit log itself is a manual record referenced across governance docs, not a system Arcane reads or writes.**
 
 ---
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "arcane-cli",
-  "version": "0.29.2",
+  "version": "0.29.3",
   "description": "Arcane framework CLI — scaffold and manage governance files in consuming repositories",
   "keywords": [
     "arcane",