Release 7.0.7 with native context and bounded recovery fixes

.claude-plugin/marketplace.base.json

@@ -9,7 +9,7 @@
   "plugins": [],
   "metadata": {
     "description": "Agent collaboration plugin marketplace",
-    "version": "7.0.6",
+    "version": "7.0.7",
     "repository": "https://github.com/sumitake/agent-collab"
   }
 }

.claude-plugin/marketplace.json

@@ -10,7 +10,7 @@
     {
       "name": "agent-collab",
       "description": "Unified collaboration package with semantic actions, source-grounded context, empirical project estimation, and a verified plugin-relative direct native runtime.",
-      "version": "7.0.6",
+      "version": "7.0.7",
       "author": {
         "name": "John Osumi"
       },
@@ -34,7 +34,7 @@
   ],
   "metadata": {
     "description": "Agent collaboration plugin marketplace",
-    "version": "7.0.6",
+    "version": "7.0.7",
     "repository": "https://github.com/sumitake/agent-collab"
   }
 }

README.md

@@ -1,11 +1,11 @@
 # agent-collab
 
 `agent-collab` publishes one collaboration plugin for Claude Code, Codex, and
-compatible hosts. Version 7.0.6 pairs a routing-only public client with the
+compatible hosts. The package pairs a routing-only public client with the
 co-packaged direct native runtime. Callers choose logical work; provider output
 remains opaque content for the calling agent to interpret.
 
-This public repository's current source is **agent-collab** (v7.0.6).
+This public repository's current source is **agent-collab** (v7.0.7).
 
 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
@@ -18,22 +18,23 @@ machine-operational contract for the repository source. The dated
 records the completed publication and keeps repository, tag, release,
 installation, and readiness claims separate.
 
-## What's new - v7.0.6
+## What's new - v7.0.7
 
-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.
+The 7.0.7 source prepares runtime `5.0.8`. Native calls keep the operating
+account's login context and SSH session markers. Bounded native output remains
+available when optional event metadata or Unicode is irregular. Eligible
+read-only Gemini work can complete one correlated empty successful turn within
+the same native process, retaining the original diagnostic and recovered answer.
 
-This content update retains the signed runtime `5.0.7` and wire schema 12
-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.
+Reviewer selection considers every contributing author family. Independent
+approval requires response-correlated native evidence; unverified results stay
+advisory. A demonstrated caller setup defect permits at most one corrected
+read-only request per original request across all descendants, without
+replaying uncertain mutations or switching providers.
+
+Signed dual-architecture delivery and staged qualification are in progress.
+The currently published release remains 7.0.6; this source preparation does not
+claim a new release, installation, or provider availability.
 
 For earlier release history, see the full [CHANGELOG](CHANGELOG.md).
 
@@ -122,18 +123,18 @@ size-branded source or generated skill surface is supported.
 ## Runtime trust boundary
 
 The canonical workspace build owns the final binary and generated manifest.
-The published package carries:
+The 7.0.7 source targets:
 
 - manifest schema 4;
 - runtime protocol 5;
 - native manifest contract 4;
-- provider runtime version `5.0.7`;
+- provider runtime version `5.0.8`;
 - one top-level closed `wire_contract` plus canonical
   `wire_contract_sha256`, bound into each artifact record; and
 - wire schema 12 with 12 logical actions and per-action timeout modes.
 
-The checked-in signed artifact rows are the imported 5.0.7 / wire-schema-12
-generation for both macOS architectures.
+Both macOS architectures carry the imported signed and notarized 5.0.8
+artifact set. Wire schema 12 is unchanged.
 
 Production provider work uses admitted progress inactivity so active work is
 not killed by a strict elapsed timer. Homogeneous `total_deadline` requests

changelog.d/2026-09-14-001-read-only-fresh-review.md

@@ -0,0 +1,9 @@
+### Changed
+
+- Document one bounded caller fresh-review for read-only review and
+  governance skills: after a completed or terminated attempt with no
+  substantive result and no uncertain external mutation, the caller may
+  issue at most one new corrected work unit to fix a demonstrated setup
+  defect while retaining the failed attempt. Shared consumed-work
+  no-replay, worker, dev-delegate, and merge-resolve contracts stay
+  unchanged.

changelog.d/2026-09-15-001-native-home-recovery.md

@@ -0,0 +1,7 @@
+### Fixed
+
+- Use the OS account's canonical home for native execution and preserve SSH
+  session markers so providers retain their own login behavior. Request files
+  remain temporary; login profiles are not copied.
+- Clarify that native recovery stays within the original invocation and that
+  existing operator authorization persists across tool steps in the same scope.

changelog.d/2026-09-15-002-reviewer-selection-evidence.md

@@ -0,0 +1,14 @@
+### Changed
+
+- Clarify that provider-free planning reports route eligibility, not reviewer
+  family identity. Candidate selection may use currently known configuration or
+  response-scoped observations; independent approval after return requires
+  response-scoped evidence distinct from the primary and every contributing
+  author family. Governance-review and plugin README invocation use the same
+  caller-verified checks. Chain keeps its existing routing exclusions; those
+  exclusions do not prove response identity or all-contributor independence.
+- Seat required initial reviewers first and reserve a tiebreaker only from spare
+  independent eligible reviewers. A sole eligible independent reviewer is used
+  initially; unmet required panel size stays visible.
+- Document optional response-scoped native model-identity observations in
+  provider-neutral public guidance without expanding the signed wire.

changelog.d/2026-09-16-review-guidance-boundaries.md

@@ -0,0 +1,6 @@
+### Fixed
+
+- Align architecture handbook guidance with all-contributor family exclusion and response-correlated native reviewer identity. Configuration may identify a candidate but cannot establish independent approval. Missing reviewer evidence leaves independent approval unmet and content advisory; it is not provider unavailability.
+
+- Keep logic-check status advisory until response-scoped evidence establishes reviewer independence, and compare every contributing author family during reviewer selection.
+- Limit the caller's corrected read-only review to one correction per original request across descendants; explicitly exclude ambiguous native mutations in generated guidance.

changelog.d/2026-09-16-runtime-508-distribution.md

@@ -0,0 +1,7 @@
+### agent-collab 7.0.7 — runtime 5.0.8
+
+- Pair runtime 5.0.8 with the public client that preserves canonical account HOME and native SSH context. Both macOS architectures must pass the coordinated signing, import and staged qualification gates before publication.
+- Gemini's existing native carrier supports one correlated read-only completion after an empty successful turn in the same process, keeping the original diagnostic and recovered answer. Codex tolerates nonprotocol startup noise and malformed optional tool metadata. Codex and ACP preserve irregular Unicode and correctly scope optional model observations. Shared interactive stdin writes remain under supervision and backpressure handling.
+- Existing native transports and descriptor-admitted actions remain in place; no provider fallback, automatic request replay, replacement login profile, broker, or new route is introduced. Claude and Grok transport selection is unchanged; no omitted Claude profile is activated.
+- Keep wire schema 12, protocol 5, native contract 4, 12 logical actions, eight logical agents and wire digest `a675807e0ff5f0544d7cc9d659914ce2dadac9be8efd0fb56635815e5c3e842a` unchanged. Bind the new public client and release-exact manifest schema to the signed producer.
+- Retain caller-owned independent-review checks, advisory results when evidence is missing, all-author reviewer selection and a nonrenewable corrected-review allowance.

docs/architecture/README.md

@@ -65,7 +65,8 @@ The public repository and package preserve these invariants:
 2. Callers select a skill or logical collaboration job. They do not receive a
    raw provider-execution escape hatch.
 3. Governance review requires a model family independent of the active primary
-   and the reviewed artifact's known author family.
+   and every contributing artifact-author family, established through native
+   identity evidence correlated to the returned review.
 4. Route authorities are closed. Read-only, output-only, and unavailable
    actions do not promote themselves because another route failed.
 5. Provider output is an artifact for the trusted primary to inspect. It does

docs/architecture/capabilities-and-workflows.md

@@ -94,16 +94,18 @@ the primary for integration.
 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.
+   candidate outside the primary and every contributing author family and bind
+   that selection to the live call. After return, verify native identity
+   evidence correlated to that response and the exact reviewed source;
+   configuration-scoped identity alone cannot establish independent approval.
 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
+This is caller-owned behavior in the [code-review
 skill](../../skill-specs/code-review.md). The routing request has no dynamic
 primary/artifact-author lineage exclusion fields.
 
@@ -175,9 +177,10 @@ A capability is usable only when all applicable gates pass:
 
 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.
+Missing independent-review evidence leaves that requirement unmet; it is not
+provider unavailability. 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).

docs/architecture/claude-participation.md

@@ -67,10 +67,11 @@ authors and integrates work, adjudicates cross-family feedback, runs
 verification, and owns landing decisions within operator authority.
 
 The native document-intent route does not change independence rules. When the
-active primary or artifact author is Anthropic-family, the primary and repository/skill workflow must exclude
-same-family evidence where an independent family is required. Conversely,
-when another family is primary, Claude document intent remains context only;
-it cannot satisfy a review or governance evidence contract.
+active primary or any contributing artifact author is Anthropic-family, the
+primary and repository/skill workflow must exclude same-family evidence where
+an independent family is required. Conversely, when another family is primary,
+Claude document intent remains context only; it cannot satisfy a review or
+governance evidence contract.
 
 ## Asynchronous participation
 

docs/architecture/governance-and-authority.md

@@ -10,22 +10,29 @@ tool call is not evidence that the caller should receive broader permissions.
 ## Independence model
 
 For governance-grade review, repository and skill policy requires a recorded
-active-primary lineage and artifact-author lineage. The primary must select
-and preserve evidence from a reviewer outside both required families. If either
-lineage is unknown, the governance workflow fails closed. OpenCode is a
-transport/host surface; the selected model's lineage supplies family provenance.
+active-primary lineage and the lineage of every contributing artifact author.
+The primary must select a reviewer outside the primary and all contributing
+author families. If any required lineage is unknown, independent approval
+remains unmet. Missing reviewer identity evidence is not provider
+unavailability: keep usable advisory content, do not stop ordinary read-only
+advisory work, and do not replay a consumed request to improve lineage.
+OpenCode is a transport/host surface; the selected model's lineage supplies
+family provenance.
 
 The current public runtime request has no primary or artifact-author-lineage
 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)
+For an authorized independent review, known native configuration or earlier
+observations can identify a candidate. The caller binds that selection to the
+actual request and honors an operator-named provider. After return, independent
+approval requires response-scoped native identity evidence correlated to that
+specific response, plus verification of the reviewed source and findings.
+Configuration alone cannot establish that approval. Planning an untargeted
+call does not reserve a 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 [review skill](../../skill-specs/code-review.md)
 contains the caller procedure.
 
 Ordinary code review remains useful when no independent reviewer is available.
@@ -41,11 +48,11 @@ cannot become governance evidence through a role assignment or successful call.
 
 ```mermaid
 flowchart LR
-    Artifact["Artifact plus author lineage"] --> Eligibility["Caller verifies required lineages"]
+    Artifact["Artifact plus all contributing author lineages"] --> Eligibility["Caller verifies required lineages"]
     Primary["Active primary lineage"] --> Eligibility
-    Eligibility --> Exclude["Caller excludes primary and author families"]
-    Exclude --> Reviewer["Caller selects an independent reviewer"]
-    Reviewer --> Evidence["Review artifact or governance verdict"]
+    Eligibility --> Exclude["Caller excludes primary and every author family"]
+    Exclude --> Reviewer["Caller selects a reviewer candidate"]
+    Reviewer --> Evidence["Caller verifies response identity, source and findings"]
     Evidence --> PrimaryGate["Primary integration and repository gates"]
 ```
 
@@ -72,7 +79,9 @@ No failure converts one row to another. In particular:
   unavailable;
 - the runtime does not itself apply output; callers must supply a disposable
   copy and verify it before application;
-- the governance workflow cannot accept an unknown primary or artifact family;
+- independent approval cannot accept an unknown primary or contributing
+  author family; that missing evidence leaves the requirement unmet and
+  does not make the provider unavailable;
 - an explicit target is not silently replaced by a different provider; and
 - a successful result does not gain merge, deployment, release, or policy
   authority.
@@ -115,8 +124,9 @@ The trace proves that required evidence was recorded in the expected form. It
 does **not** cryptographically prove that quoted review prose came from the
 claimed reviewer. This is an intentional, documented residual boundary:
 
-- skill and repository policy require family independence, verified by the
-  primary against the actual selected reviewer;
+- skill and repository policy require family independence from the primary
+  and every contributing author, verified against native identity evidence
+  correlated to the returned response;
 - repository automation validates trace form and presence; and
 - the primary, independent reviewer, and operator validate substance and exact
   head before merge.
@@ -146,11 +156,14 @@ and its evidence is verified.
 ## Why these controls matter
 
 - **Different failure modes:** independent model families are more likely to
-  expose correlated assumptions than another instance of the author's family.
+  expose correlated assumptions than another instance of the primary or a
+  contributing author's family.
 - **Least authority:** a route receives only the permission its work requires,
   reducing the cost of a hallucination or prompt-injection failure.
-- **Honest evidence:** unavailable execution and explicitly unknown lineage
-  remain uncertain; neither is converted into successful governance evidence.
+- **Honest evidence:** unavailable execution describes that attempt;
+  unknown lineage is missing independence evidence. Neither alone establishes
+  provider-wide failure or independent approval. Preserve usable advisory
+  content without claiming broader authority.
 - **Separation of duties:** authoring, reviewing, integrating, merging, and
   releasing remain distinct accountable acts.
 - **Recoverability:** an unavailable native artifact stops execution