Reconcile ACP release caveat tracker

Docs/Development/ACP_Production_Readiness.md

@@ -13,6 +13,13 @@ evidence recorded below. Remaining live-backend and host-runtime caveats are
 called out explicitly and should be resolved before release notes claim fully
 verified production deployment on a specific host.
 
+The release-caveat tracker
+[#2398](https://github.com/rmusser01/tldw_server/issues/2398) is reconciled in
+[ACP Release-Caveat Closeout - 2026-06-21](ACP_Release_Caveat_Closeout_2026_06_21.md).
+All child issues are closed; remaining sandbox, artifact, reviewer-loop,
+failure-diagnostic, and documented-only profile caveats are expected support
+boundaries rather than open tracker blockers.
+
 ## Issue Map
 
 | Issue | Workstream | Readiness role |
@@ -32,6 +39,7 @@ verified production deployment on a specific host.
 | [#1538](https://github.com/rmusser01/tldw_server/issues/1538) | ACP output artifact mapping | Maps ACP run outputs to the artifact contract and separates execution artifacts from promoted workspace work products. |
 | [#1532](https://github.com/rmusser01/tldw_server/issues/1532) | ACP-adjacent release work tracker | Tracks artifact storage/API, promotion, UI, export, verification, compatibility, and product-state follow-ups after the first ACP productionization pass. |
 | [#1704](https://github.com/rmusser01/tldw_server/issues/1704) | ACP artifact release verification | Records release-grade verification for the first accepted ACP-to-workspace-artifact golden path. |
+| [#2398](https://github.com/rmusser01/tldw_server/issues/2398) | ACP release-caveat closeout tracker | Reconciles the final child issue outcomes, surface-language consistency, and remaining expected support boundaries before closing the release-caveat tracker. |
 | [#2401](https://github.com/rmusser01/tldw_server/issues/2401) | Artifact retention and transcript redaction release policy | Makes the release retention/redaction boundaries explicit for ACP session evidence, audit records, diagnostics, artifacts, run previews, and promoted workspace artifacts. |
 | [#2400](https://github.com/rmusser01/tldw_server/issues/2400) | Sandbox host-runtime release verification | Records release-host evidence for selected ACP sandbox runtimes before sandbox-backed support claims are made. |
 | [#2402](https://github.com/rmusser01/tldw_server/issues/2402) | Live-agent caveat verification | Records deeper Goose, Hermes, and OpenCode evidence for workspace binding and non-empty MCP server injection while preserving artifact, sandbox, reviewer-loop, and failure-diagnostic caveats. |
@@ -72,6 +80,13 @@ must keep `supported_with_caveats` wording because live agent-produced ACP
 artifacts, sandbox-backed execution, reviewer-loop behavior, and failure
 diagnostic payloads remain unverified for those profiles.
 
+## Release-Caveat Closeout
+
+[#2398](https://github.com/rmusser01/tldw_server/issues/2398) is the final
+tracker for the June 2026 ACP release-caveat work. The reconciled child map and
+expected remaining support boundaries are recorded in
+[ACP Release-Caveat Closeout - 2026-06-21](ACP_Release_Caveat_Closeout_2026_06_21.md).
+
 ## Readiness Matrix
 
 | Surface | Owner modules | Required evidence | Verification commands | Pass/fail gate | Runtime caveats |

Docs/Development/ACP_Release_Caveat_Closeout_2026_06_21.md

@@ -0,0 +1,61 @@
+# ACP Release-Caveat Closeout - 2026-06-21
+
+This records the final reconciliation for
+[GitHub #2398](https://github.com/rmusser01/tldw_server/issues/2398). The
+goal is to make the release evidence boundaries explicit after the ACP
+release-caveat child workstreams closed.
+
+## Child Issue Outcomes
+
+| Issue | Outcome | Release interpretation |
+| --- | --- | --- |
+| [#2404](https://github.com/rmusser01/tldw_server/issues/2404) | Closed. Live-backend browser E2E passed across ACP Playground, Agent Registry, Agent Tasks, and Research Workspace diagnostics. | User-facing ACP flows have live-backend browser evidence when run with the bundled runner home from `Config_Files/config.txt`. |
+| [#2403](https://github.com/rmusser01/tldw_server/issues/2403) | Closed. Go runner verification passed on the recorded macOS host. | The runner build/test refresh is not a remaining release blocker. |
+| [#2401](https://github.com/rmusser01/tldw_server/issues/2401) | Closed, with [#2408](https://github.com/rmusser01/tldw_server/issues/2408) split out and closed. | Retention and redaction policy covers ACP session evidence, audit records, diagnostics, artifacts, task previews, and promoted workspace artifacts. |
+| [#2400](https://github.com/rmusser01/tldw_server/issues/2400) | Closed. Docker is the only sandbox runtime with current release-host pass evidence. | Release surfaces may claim Docker-backed sandbox runtime lifecycle evidence for the recorded host only. Lima, VZ, all-runtime sandbox support, and named-agent sandbox support remain unverified. |
+| [#2402](https://github.com/rmusser01/tldw_server/issues/2402) | Closed. Goose, Hermes, and OpenCode passed `workspace-live-e2e` on the host runner with workspace binding and non-empty MCP server injection. | These agents remain `supported_with_caveats`; artifact-producing workflows, sandbox-backed execution, reviewer-loop behavior, and failure diagnostic payloads remain unverified. |
+| [#2399](https://github.com/rmusser01/tldw_server/issues/2399) | Closed. Guardrails preserve conservative Aider, Continue, and custom-profile status. | Aider and Continue remain `documented_unverified`; the seeded `custom` profile remains template-only until a distinct named profile has live evidence. |
+
+## Surface Reconciliation
+
+- `ACP_Production_Readiness.md` now links this closeout record from the issue
+  map and status summary.
+- `ACP_Compatibility_Matrix.md` already distinguishes host E2E, workspace/MCP
+  evidence, sandbox evidence, artifact evidence, reviewer-loop evidence, and
+  failure-diagnostic evidence.
+- `tldw_Server_API/Config_Files/agents.yaml` records Goose, Hermes, and
+  OpenCode workspace/MCP evidence while keeping all three at
+  `supported_with_caveats`.
+- The setup guide and Agent Registry consume registry compatibility metadata and
+  continue to avoid promoting documented-only profiles.
+
+## Remaining Caveats
+
+These are expected support boundaries, not blockers for closing #2398:
+
+- Named downstream agents are not sandbox-supported until they pass
+  agent-specific sandbox evidence, preferably `workspace-live-e2e` with
+  `ACP_E2E_EXPECT_SANDBOX=1`.
+- Goose, Hermes, and OpenCode did not produce ACP artifacts during the
+  workspace-live runs, so artifact-producing workflows remain unverified.
+- Reviewer-loop behavior and failure diagnostic payloads remain unverified for
+  those live-agent profiles because the passing success paths did not exercise
+  those cases.
+- Aider, Continue, and seeded custom profiles remain conservative
+  documented-only entries until concrete ACP commands or named custom profiles
+  pass live certification.
+
+## Verification
+
+Closeout validation for this reconciliation slice:
+
+- GitHub GraphQL issue-state check confirmed #2398 is open and #2399, #2400,
+  #2401, #2402, #2403, #2404, and #2408 are closed.
+- Targeted `rg` audits checked ACP readiness docs, compatibility docs, setup
+  guide surfaces, seeded registry metadata, runner config, Agent Registry, and
+  ACP setup-guide fallback copy for stale caveats or overclaims.
+- Focused registry metadata tests passed for the shipped `agents.yaml` parser
+  path.
+- Bandit was run on the touched pytest file and reported the existing test-file
+  `B101` assertion baseline only. The metadata checks use ordinary pytest
+  assertions and avoid pinning the workspace-live evidence to one exact commit.

IMPLEMENTATION_PLAN_acp_final_reconciliation_2398.md

@@ -0,0 +1,22 @@
+# ACP Final Reconciliation (#2398)
+
+## Stage 1: Evidence Inventory
+
+**Goal**: Confirm all #2398 child issues and follow-ups are closed or explicitly accounted for.
+**Success Criteria**: A final child-issue map identifies the outcome for #2404, #2403, #2401, #2400, #2402, #2408, and #2399.
+**Tests**: GitHub issue state checks and local evidence-doc presence checks.
+**Status**: Complete
+
+## Stage 2: Surface Reconciliation
+
+**Goal**: Audit ACP docs, setup/registry surfaces, and compatibility language for stale caveats or overclaims.
+**Success Criteria**: Readiness, compatibility, retention/redaction, sandbox, live-agent, and setup-guide surfaces agree with final evidence state.
+**Tests**: Targeted `rg` searches over ACP docs, API setup-guide code, registry config, runner config, and frontend Agent Registry copy.
+**Status**: Complete
+
+## Stage 3: Closeout
+
+**Goal**: Apply any minimal corrections, validate, and update #2398 with a final reconciliation note.
+**Success Criteria**: Any drift is fixed, verification is recorded, and #2398 can be closed or left open with a concrete blocker.
+**Tests**: `git diff --check`, docs/search verification, and focused tests only if code surfaces change.
+**Status**: Complete

backlog/tasks/task-2396 - Finalize-ACP-release-caveat-reconciliation-for-epic-2398.md

@@ -0,0 +1,76 @@
+---
+id: TASK-2396
+title: Finalize ACP release caveat reconciliation for epic 2398
+status: In Progress
+labels:
+- ACP
+- release-closeout
+- github-2398
+references:
+- https://github.com/rmusser01/tldw_server/issues/2398
+- https://github.com/rmusser01/tldw_server/pull/2422
+modified_files:
+- IMPLEMENTATION_PLAN_acp_final_reconciliation_2398.md
+- Docs/Development/ACP_Release_Caveat_Closeout_2026_06_21.md
+- Docs/Development/ACP_Production_Readiness.md
+- tldw_Server_API/Config_Files/agents.yaml
+- tldw_Server_API/tests/Agent_Client_Protocol/test_acp_agent_registry.py
+documentation:
+- 'Implementation notes - GraphQL confirmed #2398 open and all child issues closed;
+  audit covered readiness docs plus compatibility docs plus setup guide plus seeded
+  registry plus runner config plus Agent Registry UI; corrected stale Goose Hermes
+  OpenCode registry caveats so June 20 workspace-live-e2e evidence is reflected while
+  artifact sandbox reviewer-loop and failure-diagnostic caveats remain.'
+- Verification - registry pytest passed; git diff check passed; targeted stale wording
+  search passed with the old non-empty MCP caveat only on Codex as expected; Bandit
+  was run on the touched pytest file and reported the existing B101 test assert baseline.
+- Review follow-up - removed custom assertion helpers and exact workspace-live commit
+  hash checks from the registry tests.
+- PR - https://github.com/rmusser01/tldw_server/pull/2422
+---
+
+## Description
+
+<!-- SECTION:DESCRIPTION:BEGIN -->
+
+<!-- SECTION:DESCRIPTION:END -->
+
+## Acceptance Criteria
+<!-- AC:BEGIN -->
+- [ ] #1 All #2398 child issues are confirmed closed or explicitly accounted for.
+- [ ] #2 ACP docs and setup/registry surfaces agree with final evidence state.
+- [ ] #3 Release support claims distinguish backend E2E, host stdio, sandbox, artifact, reviewer-loop, and failure-diagnostic evidence.
+- [ ] #4 Any drift is corrected with minimal changes and validation is recorded.
+- [ ] #5 #2398 receives a final reconciliation comment and can be closed when appropriate.
+<!-- AC:END -->
+
+## Implementation Plan
+
+<!-- SECTION:PLAN:BEGIN -->
+1. Audit ACP readiness, compatibility, setup, sandbox, live-agent, and retention/redaction docs against the now-closed child issue evidence.
+2. Search API/setup-guide and Agent Registry surfaces for stale release-caveat wording that conflicts with the final evidence state.
+3. Apply only minimal docs/setup corrections if drift exists.
+4. Record verification and final reconciliation evidence for #2398, then open a narrow PR or close #2398 directly if no repo changes are needed.
+<!-- SECTION:PLAN:END -->
+
+## Implementation Notes
+
+<!-- SECTION:IMPLEMENTATION_NOTES:BEGIN -->
+
+<!-- SECTION:IMPLEMENTATION_NOTES:END -->
+
+## Final Summary
+
+<!-- SECTION:FINAL_SUMMARY:BEGIN -->
+
+<!-- SECTION:FINAL_SUMMARY:END -->
+
+## Definition of Done
+<!-- DOD:BEGIN -->
+- [ ] #1 Acceptance criteria completed
+- [ ] #2 Tests or verification recorded
+- [ ] #3 Documentation updated when relevant
+- [ ] #4 Bandit run for touched code when applicable or document non-code/environment skip
+- [ ] #5 Final summary added
+- [ ] #6 Known skips or blockers documented
+<!-- DOD:END -->

tldw_Server_API/Config_Files/agents.yaml

@@ -139,7 +139,7 @@ agents:
     docs_url: "https://github.com/block/goose"
     support_state: supported_with_caveats
     verification_level: live_e2e_tested
-    compatibility_notes: "Goose native ACP completed May 23, 2026 backend live E2E through the macOS host runner on branch codex/acp-goose-backend-live-e2e commit f9ff03f88 with tldw-agent runner 0.1.0: health/setup-guide, session-new, prompt, redacted support views, diagnostics endpoint, cancel, and close passed. Sandbox, non-empty MCP injection, artifact-producing workflows, and reviewer-loop behavior remain unverified."
+    compatibility_notes: "Goose native ACP completed May 23, 2026 backend live E2E through the macOS host runner on branch codex/acp-goose-backend-live-e2e commit f9ff03f88 with tldw-agent runner 0.1.0: health/setup-guide, session-new, prompt, redacted support views, diagnostics endpoint, cancel, and close passed. June 20, 2026 workspace-live-e2e on branch codex/acp-live-agent-caveats commit ac93d96d9c verified workspace binding and non-empty MCP server injection with mcp_server_count=1. Sandbox, artifact-producing workflows, reviewer-loop behavior, and failure diagnostic payloads remain unverified."
     compatibility_docs_url: "/docs-static/Development/ACP_Compatibility_Matrix.md"
     entrypoint_strategy: native_acp
     acp_command: goose
@@ -159,7 +159,7 @@ agents:
     default: false
     support_state: supported_with_caveats
     verification_level: live_e2e_tested
-    compatibility_notes: "Hermes native ACP completed May 23, 2026 backend live E2E through the macOS host runner on branch codex/acp-hermes-live-e2e-certification commit 5e6672f8f with tldw-agent runner 0.1.0: health/setup-guide, session-new, prompt, redacted support views, diagnostics endpoint, cancel, and close passed. Sandbox, non-empty MCP injection, artifact-producing workflows, and reviewer-loop behavior remain unverified."
+    compatibility_notes: "Hermes native ACP completed May 23, 2026 backend live E2E through the macOS host runner on branch codex/acp-hermes-live-e2e-certification commit 5e6672f8f with tldw-agent runner 0.1.0: health/setup-guide, session-new, prompt, redacted support views, diagnostics endpoint, cancel, and close passed. June 20, 2026 workspace-live-e2e on branch codex/acp-live-agent-caveats commit ac93d96d9c verified workspace binding and non-empty MCP server injection with mcp_server_count=1. Sandbox, artifact-producing workflows, reviewer-loop behavior, and failure diagnostic payloads remain unverified."
     compatibility_docs_url: "/docs-static/Development/ACP_Compatibility_Matrix.md"
     entrypoint_strategy: native_acp
     acp_command: hermes
@@ -203,7 +203,7 @@ agents:
     default: false
     support_state: supported_with_caveats
     verification_level: live_e2e_tested
-    compatibility_notes: "OpenCode v1.15.7 native ACP completed May 23, 2026 backend live E2E through the macOS host runner on branch codex/acp-opencode-aider-llamacpp-certification commit 53c018269 with tldw-agent runner 0.1.0 using a local llama.cpp OpenAI-compatible server at 127.0.0.1:9099: health/setup-guide, session-new, prompt, redacted support views, diagnostics endpoint, cancel, and close passed. Sandbox, non-empty MCP injection, artifact-producing workflows, and reviewer-loop behavior remain unverified."
+    compatibility_notes: "OpenCode v1.15.7 native ACP completed May 23, 2026 backend live E2E through the macOS host runner on branch codex/acp-opencode-aider-llamacpp-certification commit 53c018269 with tldw-agent runner 0.1.0 using a local llama.cpp OpenAI-compatible server at 127.0.0.1:9099: health/setup-guide, session-new, prompt, redacted support views, diagnostics endpoint, cancel, and close passed. June 20, 2026 workspace-live-e2e on branch codex/acp-live-agent-caveats commit ac93d96d9c verified workspace binding and non-empty MCP server injection with mcp_server_count=1. Sandbox, artifact-producing workflows, reviewer-loop behavior, and failure diagnostic payloads remain unverified."
     compatibility_docs_url: "/docs-static/Development/ACP_Compatibility_Matrix.md"
     entrypoint_strategy: native_acp
     acp_command: opencode

tldw_Server_API/tests/Agent_Client_Protocol/test_acp_agent_registry.py

@@ -338,6 +338,9 @@ def test_default_agents_yaml_includes_hermes_native_acp_entrypoint() -> None:
     assert entry.acp_args == ["acp", "--accept-hooks"]
     assert entry.support_state == "supported_with_caveats"
     assert entry.verification_level == "live_e2e_tested"
+    assert "workspace-live-e2e" in entry.compatibility_notes
+    assert "non-empty MCP server injection" in entry.compatibility_notes
+    assert "non-empty MCP injection, artifact-producing workflows" not in entry.compatibility_notes
 
 
 def test_default_agents_yaml_includes_goose_backend_live_e2e_metadata() -> None:
@@ -362,6 +365,9 @@ def test_default_agents_yaml_includes_goose_backend_live_e2e_metadata() -> None:
     assert entry.verification_level == "live_e2e_tested"
     assert "backend live E2E" in entry.compatibility_notes
     assert "commit f9ff03f88" in entry.compatibility_notes
+    assert "workspace-live-e2e" in entry.compatibility_notes
+    assert "non-empty MCP server injection" in entry.compatibility_notes
+    assert "non-empty MCP injection, artifact-producing workflows" not in entry.compatibility_notes
 
 
 def test_default_agents_yaml_includes_opencode_backend_live_e2e_metadata() -> None:
@@ -387,6 +393,9 @@ def test_default_agents_yaml_includes_opencode_backend_live_e2e_metadata() -> No
     assert "backend live E2E" in entry.compatibility_notes
     assert "local llama.cpp" in entry.compatibility_notes
     assert "commit 53c018269" in entry.compatibility_notes
+    assert "workspace-live-e2e" in entry.compatibility_notes
+    assert "non-empty MCP server injection" in entry.compatibility_notes
+    assert "non-empty MCP injection, artifact-producing workflows" not in entry.compatibility_notes
 
 
 def test_default_agents_yaml_includes_codex_backend_live_e2e_metadata() -> None: