sumitake · sumitake · Sep 8, 2026 · Sep 8, 2026
@@ -7,7 +7,7 @@ remains opaque content for the calling agent to interpret.
 
 This public repository's current source is **agent-collab** (v7.0.6).
 
-Current published release: **7.0.5** ([`v7.0.5`](https://github.com/sumitake/agent-collab/releases/tag/v7.0.5)).
+Current published release: **7.0.6** ([`v7.0.6`](https://github.com/sumitake/agent-collab/releases/tag/v7.0.6)).
 It carries signed provider runtime `5.0.7`. Host installation, readiness, and
 provider availability remain separate evidence planes.
 
@@ -20,17 +20,20 @@ installation, and readiness claims separate.
 
 ## What's new - v7.0.6
 
-Review skills now make reviewer independence a caller responsibility. Before
-dispatch, the caller establishes the primary and artifact-author families and
-checks the proposed reviewer. Afterward, it verifies the observed reviewer
+When independent review is required, the skills make reviewer independence a
+caller responsibility. Before dispatch, the caller establishes the primary and
+artifact-author families and checks the proposed reviewer. Afterward, it verifies the observed reviewer
 family and source before accepting independent evidence. Unknown lineage stays
 unknown; a routing result alone does not establish independence.
 Ordinary code review can still use available Gemini when no eligible independent
 reviewer exists, with an explicit advisory label and no independent-approval claim.
 
 This content update retains the signed runtime `5.0.7` and wire schema 12
-unchanged. Publication and host refresh for 7.0.6 are pending; the dated status
-snapshot remains the verified 7.0.5 release record.
+unchanged. The release is published, and all 178 released files match each of
+the four observed local installations. Provider-free planning and migration
+checks passed from the control session against those roots; existing tasks
+still need a fresh session to load updated skills. The dated status snapshot
+records those limits and the unresolved Claude subscription-access denial.
 
 For earlier release history, see the full [CHANGELOG](CHANGELOG.md).
 
@@ -189,7 +192,7 @@ Dependabot for dependency update review.
 
 After every other release task finishes, complete the
 [documentation closeout](docs/architecture/repository-and-release.md#final-documentation-closeout).
-The v7.0.5 closeout is recorded in the
+The v7.0.6 closeout is recorded in the
 [status and evidence snapshot](docs/architecture/status-and-evidence.md). Each
 future closeout must likewise align the architecture handbook, this README,
 and generated changelog evidence with the exact release without exposing

@@ -0,0 +1,4 @@
+### Changed
+
+- Record the published agent-collab 7.0.6 release and verified assets, separating installation and provider-free readiness from fresh-session loading and Claude inference access.
+- Align the handbook and release narratives with caller-owned independent review, ordinary Gemini advisory review, bounded source delegation and compatible caller sandbox execution. The signed runtime and release history remain unchanged.
@@ -10,7 +10,7 @@ the contribution contract remain authoritative for their respective claims.
 When two generations differ, use the status vocabulary below instead of
 silently treating them as the same thing.
 
-The current published release is v7.0.5 with signed provider runtime 5.0.7,
+The current published release is v7.0.6 with signed provider runtime 5.0.7,
 manifest schema 4, runtime protocol 5, native contract 4, and wire schema 12.
 
 ## Read this handbook in order
@@ -105,7 +105,7 @@ not a public issue or pull request.
 | --- | --- | --- |
 | What does the package install? | [System context](system-context.md) | Host manifests, generated marketplaces, package tree, and distribution tests. |
 | Which workflows exist? | [Capabilities and workflows](capabilities-and-workflows.md) | `skill-specs/`, generated skills, package reference, and skill-contract tests. |
-| How are agent projects estimated? | [Project estimation](project-estimation.md) | Public request/result schemas, deterministic helper, released v7.0.5 maintenance evidence, skill checkpoints, maintenance verifier, producer-byte compatibility fixtures, and focused tests. |
+| How are agent projects estimated? | [Project estimation](project-estimation.md) | Public request/result schemas, deterministic helper, released v7.0.6 maintenance evidence, skill checkpoints, maintenance verifier, producer-byte compatibility fixtures, and focused tests. |
 | Who may review, write, or merge? | [Governance and authority](governance-and-authority.md) | Host policy, coordinator, public governance contract, PR template, and compliance checks. |
 | Is a route usable now? | [Status and evidence](status-and-evidence.md) | Installed version plus provider-free readiness on that host. Repository presence alone is insufficient. |
 | How do I install or recover? | [Lifecycle and operations](lifecycle-and-operations.md) | Current host CLI, migration doctor, runtime-management surface, and migration policy. |

@@ -20,7 +20,7 @@ organizes that inventory by user outcome.
 | Planning, architecture, and estimation | `architect`, `brainstorm`, `compose-skills`, `intent-check`, `project-estimation`, `second-opinion` | Clarify a design, widen options, select workflows, obtain an independent read, or forecast agent-led delivery. |
 | Governance and assurance | `autonomy-readiness`, `code-review`, `governance-review`, `logic-check`, `qa-verify`, `red-team`, `untrusted-audit` | Test correctness, independence, security, provenance, and completion evidence. |
 | Deliberation and stakeholder lenses | `debate`, `simulate-user` | Expose conflicting arguments or test a proposal against a persona. |
-| Delegation and implementation | `delegate`, `dev-delegate`, `worker` | Return bounded research or output for the primary to integrate. |
+| Delegation and implementation | `delegate`, `dev-delegate`, `worker` | Analyze supplied sources or return a development artifact for the primary to integrate. |
 | Context and knowledge work | `context`, `knowledge-compile`, `project-knowledge` | Extract or synthesize bounded documents/repositories, or maintain an explicit project knowledge layer, with provenance. |
 | Reproducible workflows | `chain`, `chain-configurator`, `orchestrate` | Define and execute repeatable multi-step coordination. |
 | Integration and conflict handling | `merge-resolve` | Analyze and resolve a bounded merge conflict while preserving intent. |
@@ -86,34 +86,43 @@ the primary for integration.
 
 ## Common workflows
 
-### Independent review
+### Code review and independent approval
 
-1. Identify the exact artifact and its author lineage.
-2. Invoke `second-opinion`, `code-review`, `governance-review`, `red-team`, or
-   `qa-verify` according to the risk.
-3. The skill/repository workflow requires a reviewer outside the primary and
-   artifact-author families. The primary verifies that independence; the
-   current routing request does not dynamically enforce those lineages.
-4. Preserve the raw finding and typed status.
-5. The primary adjudicates, changes the artifact if needed, and re-verifies the
-   exact head.
+1. Identify the artifact, primary, authors and whether the task requires
+   independent approval.
+2. Use the appropriate review skill. For required independence, verify a
+   reviewer outside the primary and author families and bind that selection to
+   the live call. Verify observed reviewer lineage and exact source afterward.
+3. For ordinary code review, an available Gemini reviewer can still help when
+   no distinct-family reviewer is available. Label same-family or unknown
+   lineage as advisory; leave any independent approval requirement unmet.
+4. Preserve and adjudicate the raw findings. Do not infer independence from
+   routing, role names or a subscription, repeatedly attempt an unavailable
+   provider, or replay a consumed request to repair formatting or evidence.
+
+This is caller-owned behavior in the [released code-review
+skill](../../skill-specs/code-review.md). The routing request has no dynamic
+primary/artifact-author lineage exclusion fields.
 
 ### Bounded delegation
 
-1. The primary keeps objective interpretation and integration ownership.
-2. A worker receives a bounded scope, authority, evidence contract, and stop
-   condition.
-3. Read-heavy work can use a cheaper capable tier; judgment and landing stay
-   with the primary.
-4. The caller supplies a disposable repository or copy, captures any patch,
-   and verifies it before application; the runtime does not apply output.
-5. The primary reviews and tests all delegated output before it becomes source.
+The primary keeps objective interpretation and integration ownership. Ordinary
+`delegate` work analyzes supplied bounded documents or an exact sealed
+repository through an admitted context action. A list of names, links or topics
+alone is not a document corpus, and this route does not promise source discovery.
+Each worker receives the relevant sources, scope and stop condition. Results
+remain advisory; multiple workers do not imply multiple model families.
+
+Development delegation uses the separate admitted code-generation actions.
+The caller supplies a disposable repository or copy, captures any returned
+patch, and reviews and tests it before application. A worker does not gain
+permission to apply output or make landing decisions through successful routing.
 
 ### Architecture and planning
 
-Use `brainstorm` to widen the option space, `architect` for an independent
+Use `brainstorm` to widen the option space, `architect` for an additional
 read-only architecture consultation, `architecture-review` for a primary-led
-codebase sweep, `intent-check` to validate task interpretation, and
+codebase sweep, `intent-check` for an advisory task-interpretation comparison, and
 `decision-map` when the effort is too large for one session.
 
 `project-estimation` adds a deterministic delivery forecast once a formal
@@ -124,7 +133,7 @@ workflows explicitly compose the checkpoint before final presentation. A host
 without contextual skill selection uses explicit invocation and reports that
 the automatic checkpoint was unavailable. See
 [Project estimation](project-estimation.md) for modes, examples, output
-semantics, and the published v7.0.5 maintenance evidence.
+semantics, and the published v7.0.6 maintenance evidence.
 
 The packaged prior is currently an explicit bootstrap: enhancement duration is
 descriptive, bootstrap confidence cannot be high, and unsupported greenfield,
@@ -145,7 +154,7 @@ authority.
 | --- | --- | --- |
 | Claude Code | Claude-compatible plugin manifest and marketplace metadata. | Native package install and `/agent-collab:*` skills. The official native CLI may serve read-only document intent when action-scoped readiness passes; Claude is not eligible for managed review, governance, repository, or code-generation actions (see [Claude participation](claude-participation.md)). |
 | Codex CLI/app | Codex-native manifest and generated Codex marketplace. | Native package install and the same skill namespace. Start a new task after install/update. |
-| Antigravity | Dynamic host policy and async target model; no separate package. | Use only through a compatible package host and observed async readiness. It is not a coordinator `target_agent` unless the current signed descriptor explicitly admits it. |
+| Antigravity | Compatible plugin import, logical Gemini managed routes, and separate host-owned async coordination; no separate package. | Gemini repository review uses the co-packaged coordinator and action-scoped readiness. Async readiness is a different surface; neither host name nor reviewer role proves independence. |
 | OpenCode and ZCode | Dynamic host/model policy and managed OpenCode routes; no separate package. | A compatible host/plugin surface is required. OpenCode is a transport; the selected model supplies family lineage. |
 | Custom host | Explicit primary identity fields and the closed package contract. | If the host cannot load the package safely, it is unsupported; do not recreate provider-specific shims. |
 
@@ -161,9 +170,11 @@ A capability is usable only when all applicable gates pass:
 - the native boundary and provider-free readiness checks pass; and
 - provider authentication, quota, and request execution succeed.
 
-Failing one gate produces a typed result such as `unavailable`,
-`same_family_blocked`, `unknown_family`, or a runtime error category. Do not
-turn that result into a raw-provider fallback or broader authority.
+Runtime and planning diagnostics describe the attempted route; they do not
+perform the caller's family-exclusion check or establish provider-wide failure.
+Missing independent-review evidence leaves that requirement unmet. Preserve
+available advisory content without claiming broader authority, silently
+substituting an operator-named target, or replaying a consumed request.
 
 For installation and recovery, continue to
 [Lifecycle and operations](lifecycle-and-operations.md).
@@ -52,6 +52,12 @@ The 7.0.5 host qualification explicitly deferred restoration of local Claude
 subscription access and live Claude intent qualification. That deferral does
 not enable another Claude action and is not a passed native inference check.
 
+During the 7.0.6 follow-up, one fresh admitted document-intent attempt returned
+an organization subscription-access denial. Package installation and
+provider-free readiness did not establish inference access. Omitted Claude
+profiles remain unqualified in [#162](https://github.com/sumitake/agent-collab/issues/162);
+the attempt was not replayed, and no billing or credential change was made.
+
 ## Host and resident-primary role
 
 Claude Code remains a fully supported host: the package installs natively,

@@ -20,12 +20,31 @@ field. It does not dynamically perform this exclusion. The skill and repository
 workflow require it, and the primary verifies the selected reviewer and the
 substance of the exact-head review.
 
+For an authorized independent review, the caller binds its verified reviewer
+selection to the actual request, honors an operator-named provider, and checks
+observed native lineage afterward. Planning an untargeted call does not reserve
+that selection for a later call. Role names, receipts and routing success do
+not establish independence, and a consumed request is not replayed to improve
+its evidence. The [released review skill](../../skill-specs/code-review.md)
+contains the caller procedure.
+
+Ordinary code review remains useful when no independent reviewer is available.
+For example, a Gemini primary with only Gemini available can receive a
+same-family advisory review. Unknown lineage is labeled unverified advisory.
+Neither clears an independent approval required by the task or repository, and
+the caller does not repeatedly try an absent Grok installation. An OpenCode or
+ZCode subscription name alone establishes no model lineage.
+
+Intent comparison, creative ideation, persona simulation and bounded context
+delegation are advisory work. They need no family difference to proceed and
+cannot become governance evidence through a role assignment or successful call.
+
 ```mermaid
 flowchart LR
-    Artifact["Artifact plus author lineage"] --> Eligibility["Eligibility policy"]
+    Artifact["Artifact plus author lineage"] --> Eligibility["Caller verifies required lineages"]
     Primary["Active primary lineage"] --> Eligibility
-    Eligibility --> Exclude["Exclude primary and author families"]
-    Exclude --> Reviewer["Eligible independent reviewer"]
+    Eligibility --> Exclude["Caller excludes primary and author families"]
+    Exclude --> Reviewer["Caller selects an independent reviewer"]
     Reviewer --> Evidence["Review artifact or governance verdict"]
     Evidence --> PrimaryGate["Primary integration and repository gates"]
 ```

@@ -72,7 +72,7 @@ Installation, selection, and readiness are separate checks.
    snapshot. A listed route is not proof of readiness.
 5. **Bounded smoke test:** invoke one low-risk read-only workflow, such as a
    second opinion on a short draft, and verify the returned family is eligible
-   and independent.
+   and independent where the task requires independent approval.
 
 For an activation package, inspect the provider-free migration report:
 
@@ -99,7 +99,7 @@ Invoke the skills in normal language or by their host command. Examples:
 /agent-collab:second-opinion Review this architecture decision.
 /agent-collab:code-review Review the current diff against the task.
 /agent-collab:qa-verify Verify the completed work against these acceptance criteria.
-/agent-collab:delegate Split this read-only research list with an independent reviewer.
+/agent-collab:delegate Extract comparable facts from these supplied reports in parallel.
 /agent-collab:agent-runtime-status
 ```
 
@@ -119,7 +119,7 @@ Preserve the typed coordinator response. Any later issue report is an explicit,
 separately authorized action and never grants governance authority or licenses
 replay of the provider request.
 
-`project-estimation` is offline and read-only by default. The packaged v7.0.5
+`project-estimation` is offline and read-only by default. The packaged v7.0.6
 source contains an explicit bootstrap prior: enhancement duration is
 descriptive, greenfield may return `no_compatible_prior`, and absent token
 evidence returns `unavailable_no_token_prior` rather than zero. Persist an
@@ -217,14 +217,26 @@ a fresh session only after the host reports the intended package version. Do
 not reconstruct a lifecycle path, copy a runtime out of another package, or
 fall back to a retired provider-specific plugin.
 
+## Caller sandbox compatibility
+
+A caller's macOS sandbox can prevent a native provider from initializing its
+own command sandbox, even though file reads work. Once this composition problem
+is established, use the host's approved execution path for the normal
+coordinator while preserving native sandboxing and permission decisions. Treat
+it as a caller invocation constraint, not a provider failure. If approval is
+unavailable, stop the dependent action; do not disable native protections or
+replay a consumed request under different permissions. The
+[package reference](../../plugins/agent-collab/README.md#routing-request)
+contains the invocation guidance.
+
 ## Troubleshoot
 
 | Symptom or status | Meaning | Safe response |
 | --- | --- | --- |
 | Skill is missing | The package may not be installed, enabled, or loaded in this session. | Check host plugin inventory, then start a new session/task. |
 | `duplicate_blocked` or migration conflict | A retired package remains active or installed. | Run migration doctor, apply only its host-specific removal actions, and run it again. |
 | `unavailable` | The route, runtime, provider prerequisite, or observed readiness is not currently usable. | Run runtime status and migration doctor; check supported vendor authentication separately. Do not use a raw-provider fallback. |
-| Reviewer shares a required excluded family | The primary cannot count that output as independent governance evidence. | Select an eligible different family before dispatch or record the existing review as non-independent; the current routing wire does not enforce lineage exclusion. |
+| Reviewer shares a required excluded family | The primary cannot count that output as independent governance evidence. | Where independence is required, select an eligible different family before dispatch. Ordinary review may retain the output as advisory; the current routing wire does not enforce lineage exclusion. |
 | Reviewer or author lineage is unknown | The primary cannot establish governance independence. | Verify the actual model lineage and authorship; do not guess from a host nickname or installation path. |
 | `config_error` | Request fields, host identity, or route/action pairing violate the closed schema. | Use the installed skill/package reference; remove unsupported fields rather than widening the schema. |
 | `auth_error` or `quota_error` | The managed provider prerequisite failed after routing. | Use the provider's supported login/account interface or wait for quota. Keep the same authority. |