Repository navigation
feat: integrate Pool native ACP Chat with scoped lifecycle proof - #8201
milesatao-svg wants to merge 29 commits into
Conversation
Preserve harness-owned evidence and merge current canonical API, storage, and native Chat updates. Runtime proof remains qualified by its original source.
Build the real daemon in PR CI for isolated VPS runtime verification without local validation suites or dependency installations. Record source revision, Go build info, and executable hash.
Copied source evidence 1/2Historical material from original PR #6551, copied for this account migration. Source PR head when copied: Original comment by @nikhilachale on 2026-10-11T16:25:31ZPool qualification is BLOCKED; this draft is unregistered. Official native Pool1.0.16 initialized ACP and created sessions, but the available authorized Z.ai coding endpoint failed native model turns with404 at Immutable native hold report and curated qualification record are published in PR #6509. Both were downloaded and SHA-256 verified after publication. Native executable hash is Actual AO Chat/API/nativeacp, permissions/cancellation/replay and Electron screenshot gates are NOT_RUN: no successful native provider turn enabled safe production registration, and no AO session was launched. Screenshot unavailable for that reason; no synthetic image substitutes. Three exact owned native PIDs are absent after bounded cleanup; production AO and the preparation checkout remain untouched. Source head Original comment by @nikhilachale on 2026-10-11T16:38:21ZFinal Pool stage result: BLOCKED on a native provider prerequisite; PR #6551 remains draft and unregistered.
Native Pool1.0.16 initializes/new-sessions successfully and hard-rejects unknown history, but its first model turn emits No production registration, migration/API/UI, reviewer or handoff was added. Native tool/private-instruction/permission/cancel/replay and actual AO Chat/API/host/Kill/Electron screenshot gates are NOT_RUN; screenshot unavailable because no safely admitted AO session exists. No substitute pixels or integration claim. All three owned native PIDs are absent; production PID625660 remains present, native profiles/history/preparation and old evidence are preserved. Only the hashed, regenerable release archive was removed after verifying no open references; the executable remains available. No merge, release or deployment. Original comment by @nikhilachale on 2026-10-11T21:20:15ZPool recovery is now implemented in the existing branch. Immutable report and process/provenance receipts. Backend/UI head PNG SHA256: active Final exact-head CI: source backend/race gate, source frontend, evidence frontend, 150/150 fixtures. Stage complete; no merge or release. Original comment by @nikhilachale on 2026-10-12T16:55:26ZHarness-specific reports, JSON receipts, screenshots and provenance now live on this integration PR branch, rather than the skill PR #6509. Evidence index with embedded screenshots and JSON links. Historical successes, failures and scope limitations remain as recorded. Moving these files does not rerun native/AO testing or upgrade any verdict. This is evidence relocation only. Existing tested source/native hashes and failures are preserved; the documentation commit is not a fresh runtime pass. Applicable checks run in PR CI. Original comment by @nikhilachale on 2026-10-12T18:08:39ZCanonical main conflicts reconciled at Validation is PR CI only; current-head checks are pending until GitHub reports them. This merge does not constitute a new native/AO run or promote historical fresh/restore/provider-role proof. Harness evidence stays on this integration PR; PR6509 remains reusable guidance and linked summaries. Original comment by @nikhilachale on 2026-10-12T18:18:07ZCanonical main conflicts reconciled at Validation is PR CI only; current-head checks are pending until GitHub reports them. This merge does not constitute a new native/AO run or promote historical fresh/restore/provider-role proof. Harness evidence stays on this integration PR; PR6509 remains reusable guidance and linked summaries. Original comment by @nikhilachale on 2026-10-12T20:59:37ZCanonical main conflicts reconciled at Validation is PR CI only; current-head checks are pending until GitHub reports them. This merge does not constitute a new native/AO run or promote historical fresh/restore/provider-role proof. Harness evidence stays on this integration PR; PR6509 remains reusable guidance and linked summaries. Original comment by @nikhilachale on 2026-10-13T00:00:02ZCanonical main and the green shared ACP replay CI budget reconciled at The shared test-only fix PR6617 had21 green applicable checks; this integration head still needs its own exact-head CI. The full20543-frame replay burst and exact-once assertions remain unchanged. No routine local suites ran. This merge does not constitute a new native/AO run or promote historical fresh/restore/provider-role proof. Harness evidence stays on this integration PR; PR6509 remains reusable guidance and linked summaries. Original comment by @nikhilachale on 2026-10-21T05:00:54ZFresh testing-only retest starting for Original comment by @nikhilachale on 2026-10-21T05:11:31ZCurrent bounded testing receipt — PR #6551 — BLOCKED / integration incomplete Tested source/current head: This is fresh Linux x86_64 proof. Historical committed notes/screenshots are references only. Actual current frontend never ran: its node_modules is absent, and installation/symlinking was prohibited. AO UI evidence: BLOCKED. AO UI attachment evidence: BLOCKED. No working approved GitHub attachment uploader; no PNG was produced or published, and no screenshot checkpoint counts as attached. Actual provenance:
Go: 1.27.1; Node: v24.21.0. Minimal live-only backend build ran serially with GOMAXPROCS=1, GOFLAGS=-p=1, GOMEMLIMIT=384MiB, Node heap 384MiB, existing caches; start 05:02:55.169677Z, finish 05:03:40.016383Z, exit 0. Available disk 2229 MiB before build and 2063 MiB at final inventory; no broad cleanup. No local unit/race/lint/vet/typecheck/frontend/fixture suites ran (CI ONLY). Distinct actual attempts and first failures:
Native proof file SHA-256: Complete completion matrix:
Complete screenshot/action/result checkpoint index (all captures and attachments unavailable; API/source observations never substitute for images):
Current-head CI readback (all listed checks have exact head 5667039): 24 check-runs completed success, zero failing/pending/unavailable returned check-runs. Legacy combined status returned pending with total_count=0 and no contexts; no legacy pending job exists to name. This does not prove live provider/UI conformance.
Reproduction/current-status receipt: source and skill are pinned above. Safe generic reproduction uses existing authorized native credentials in the environment (never in argv/reports) and a new isolated RUNTIME directory: GOMAXPROCS=1 GOFLAGS=-p=1 GOMEMLIMIT=384MiB go build -o RUNTIME/bin/ao ./cmd/ao
pool --version
pool acp --help
# Native proof used ACP initialize, session/new, session/set_mode(always-allow),
# session/prompt with a single disposable workspace/file witness; then exact cleanup.
TMPDIR=RUNTIME/tmp node SKILL/scripts/audit_ao.mjs --agent pool \
--repo SOURCE --ao RUNTIME/bin/ao --local-contract RUNTIME/contract.json \
--report RUNTIME/NEW_ATTEMPT/report --timeout-seconds 25 \
--startup-timeout-seconds 35 --keep-daemonActual first-strict contract (binary path generalized here; no credentials): {"schemaVersion":1,"agents":{"pool":{"binary":"INSTALLED_POOL","versionArgs":["--version"],"integration":{"args":["acp","--version"],"expect":"1.0.16"}}}}Exact private contract byte hash: Source corroboration: current Pool guide; Chat binding/permission gates; TUI rejection before allocation; actual Harness settings controls; auth-plan registry; separate reviewer vocabulary. These links describe implementation, not runtime proof. Cleanup PASS: native PID 1449208/kernel-start 160478441 was settled, then stopped with bounded TERM/KILL (exit -9); PID absent at 05:04:07.613078Z. No active command was witnessed, so this is not cancellation proof. Owned AO PID 1449284/kernel-start 160481450 had zero task sessions; verified executable/readyz identity before POST /shutdown, HTTP 202 at 2026-10-21T05:09:35.865517+00:00; exact PID absent by 2026-10-21T05:09:36.149194+00:00, remainingOwnedPids=[]. No UI process was launched. Source clean, no untracked evidence. Parent still performs final process verification and removes only its clean detached source worktree. Retained privately outside every Git checkout, under this PR's sibling runtime only: plan/matrix/checkpoint index, exact contract, current build/native binaries or reference link, isolated native/AO state/history, safe receipts, bounded logs, first strict report/issues, and empty attachment-queue.json. Profiles/state/DBs/raw logs are not public attachments. No credential-file copy or entire store copy was made; no original native profile changed. No evidence commits, fake screenshots/URLs, production daemon action, installations/upgrades, login reset or reviewer publication. Unresolved/next actions: keep integration incomplete; obtain explicit approval before any configured-only diagnostic AO Chat action; qualify registered Chat through actual shared UI including streaming/history/approve/reject/cancel/recovery/exact restore/instructions; safely provide current-head Electron dependencies and approved image attachment path; qualify both model surfaces/persisted visibility and provider/account controls; assess missing AO login and reviewer wiring separately; resolve PR conflicts in an authorized implementation worker. PR-description evidence link is pending parent update; this worker does not edit it. Green CI, native success and worker exit do not establish completion. Final pre-publication head readback: Original comment by @nikhilachale on 2026-10-21T05:37:12ZExplicitly approved isolated AO Chat diagnostic — BLOCKED User approval was given on 2026-10-21: “Run isolated AO Chat diagnostic”, using the same already-tested DeepSeek credential. Prior automatic-review block is preserved: #6551 (comment). One registered Chat session and at most one provider turn; no model refresh, account/config changes, installs, source changes or production restart. Source: 5667039; head at publication: 5667039 Diagnostic provider/file result: BLOCKED; actual sessions: 0; requested turns: 1; persisted Chat mode: False; file token/cwd witness: False; proof SHA256: unavailable; native identity digest: unavailable. Reason/limits: registered Chat spawn rejected; preserved API envelope; no retry
Strict auth remains configured/unverified and strict TUI audit remains BLOCKED. This diagnostic does not qualify real UI screenshots/attachments, full Chat journeys, cancellation, exact restore, reviewer, models/provider controls or onboarding. No existing evidence was edited. Fresh isolated readyz PID/executable/data ownership: PASS. Cleanup: PASS; remaining owned process count 0; compute lease held 2.385 seconds. Original comment by @nikhilachale on 2026-10-21T05:49:09ZExplicitly approved isolated AO Chat diagnostic — FAIL User approval was given on 2026-10-21: “Run isolated AO Chat diagnostic”, using the same already-tested DeepSeek credential. Prior automatic-review block is preserved: #6551 (comment). One registered Chat session and at most one provider turn; no model refresh, account/config changes, installs, source changes or production restart. Source: 5667039; head at publication: 5667039 Diagnostic provider/file result: FAIL; actual sessions: 1; requested turns: 1; persisted Chat mode: False; file token/cwd witness: False; proof SHA256: unavailable; native identity digest: unavailable. Reason/limits: 'workspacePath'
Strict auth remains configured/unverified and strict TUI audit remains BLOCKED. This diagnostic does not qualify real UI screenshots/attachments, full Chat journeys, cancellation, exact restore, reviewer, models/provider controls or onboarding. No existing evidence was edited. Fresh isolated readyz PID/executable/data ownership: PASS. Cleanup: PASS; remaining owned process count 0; compute lease held 5.807 seconds. Original comment by @nikhilachale on 2026-10-21T05:51:32ZPool approved Chat diagnostic — corrective read-only evidence receipt The user explicitly approved “Run isolated AO Chat diagnostic” on 2026-10-21 using the same previously tested DeepSeek credential. This corrects the operator observer failure, not the provider/authentication result. Preserve both prior attempts: original automatic-review block, scratch default-branch setup failure, and observer-error attempt. None was edited. Tested/current PR head:
The provider assistant text was: “I'll check the working directory and write the file.” Final durable turn state is Lease acquired Outcome: registered Chat/provider streaming/tool transport observed; complete provider/file test interrupted by operator observation error. No full diagnostic/integration PASS and no auth-state upgrade. This receipt used only retained isolated SQLite/files and GitHub metadata; no additional live work was run. Original comment by @nikhilachale on 2026-10-21T06:34:03ZAuthorized implementation repair (repair) starts at PR head 5667039, latest main 0c11049, readonly harness skill a67b5d2. Pool/Docker existing branch fixes only; surgical code and regression tests are allowed. Actual worker gpt-6.1-sol/high/fast, 90-minute maximum. Routine suites run in CI only; bounded live/build/UI work uses the shared lease and full proof waits for the existing retest queue to finish. Old evidence/comments and production remain preserved. Original comment by @nikhilachale on 2026-10-21T06:39:45ZImplementation repair started on the existing Pool PR. Freshly fetched PR head is Reassessment found shared model visibility controls already on main. The scoped repairs are native Pool Account setup/logout wiring, keeping setup reachable when auth is unknown, and avoiding invented provider groups from model namespace text. Auth remains configured/unknown without a safe verification result; inventory/model discovery will not send a test prompt. The earlier disposable default-branch fixture error and subsequent workspace observer error are preserved as test-plumbing failures. Historical native file proof and registered Chat streaming/pwd remain distinct; they do not establish current-head qualification or verified AO authentication. Routine build/test/race/lint/typecheck/fixture suites are CI-only under this repair policy. Any bounded runtime work uses the shared non-waiting lease. Long native/AO lifecycle and actual Electron proof are deferred to the final-proof phase if compute/capacity does not permit them. Actual UI capture and published screenshot attachments remain BLOCKED; screenshot unavailable because no current UI run or approved uploader exists in this repair phase. TUI, Reviewer and handoff remain outside this Chat-only scope. Original comment by @nikhilachale on 2026-10-21T06:46:01ZImplementation repair (repair) incomplete at 2026-10-21T06:46:00.347268+00:00. free capacity fell below 700MiB. Owned cleanup: PASS. Worktree retained at its private operational location, including uncommitted work. No merge/deploy/force-push occurred. Current source/CI/proof claims are only those explicitly recorded by the worker; missing proof and durable UI attachments remain gaps. Original comment by @nikhilachale on 2026-10-21T07:27:35ZAuthorized implementation repair (final-proof) starts at PR head 5667039, latest main 474962e, readonly harness skill 2bca6e7. Pool/Docker existing branch fixes only; surgical code and regression tests are allowed. Actual worker gpt-6.1-sol/high/fast, 90-minute maximum. Routine suites run in CI only; bounded live/build/UI work uses the shared lease and full proof waits for the existing retest queue to finish. Old evidence/comments and production remain preserved. |
Copied source evidence 2/2Historical material from original PR #6551, copied for this account migration. Source PR head when copied: Original comment by @nikhilachale on 2026-10-21T07:40:55ZPool final-proof repair is pushed at Implemented and recovered from the preserved unfinished repair: native Account setup/logout plans, reachable setup while auth is unknown, provider grouping only from native metadata, explicit Chat-only terminal refresh refusal, and regression source for those boundaries. Main already supplies shared model visibility controls; no duplicate model subsystem was added. The unmerged Pool migration now uses the repository timestamp generator above main, with updated inverse/legacy-schema regression source. API artifacts were regenerated from Go sources. Required API generation and one minimal serial AO build passed under the shared lease. The first build attempt failed because the isolated sparse checkout omitted a Go workspace module; that original failure is retained, then corrected before retry. No provider call was made during generation/build. Routine unit/race/lint/vet/typecheck/fixture suites are CI-only for this task; fresh current-head checks are now being inspected. Authentication remains configured/unknown: no safe native authorization-status verifier is wired, and neither inventory nor model discovery sends a test prompt. Supported Chat execution will be qualified separately with an explicit disposable task. Native, registered AO Chat, and actual Electron proof are independent pending rows. TUI, Reviewer and handoff remain intentionally excluded. Initials branding remains the documented fallback without a verified reusable logo license. The read-only hygiene scan passes formatting, whitespace, migration numbering and main freshness. It still flags the original PR's historical evidence files/host paths and pre-existing proof workflow; these originals are preserved unchanged, and no new report, log or screenshot is committed. Scanner misses caused by current registry/docs layouts are not treated as runtime failures or passes. Actual current-source Electron UI proof is still pending a capacity forecast. Published screenshot attachments remain BLOCKED because no approved working uploader is available. This is an incomplete integration qualification, not a full PASS. Original comment by @nikhilachale on 2026-10-21T07:48:41ZFresh Pool native + registered AO Chat functional proof: 19 recorded checks PASS, with auth/UI qualification gaps retained. Tested clean source
There were 11 explicit provider turns total across native and AO proof. Scratch Git setup used an owned main commit, a local origin and cached Functional proof receipt SHA-256 Owned cleanup: PASS, zero survivors verified by PID plus kernel birth; shared compute held 136.845 seconds. No production state, original profiles, older evidence or historical comments were changed. The original sparse-module build failure is retained separately from the corrected successful build/live attempt. Limits remain explicit: safe AO authorization verification is unavailable; no login/logout action was performed, no unrelated live control-session isolation, daemon replacement, negative-history restore or actual current-source Electron journey is established by this attempt. Reviewer/handoff remain excluded and need their honest scope disposition separately. Model visibility UI/onboarding/provider-maintenance UI are not covered by API/native results. Actual UI capacity is being measured. Screenshot publication remains BLOCKED without an approved uploader. This does not establish full integration PASS. Original comment by @nikhilachale on 2026-10-21T08:06:42ZCurrent-head Pool functional evidence — The first pushed repair's SQLite CI failed because Pool's new schema check masked main's existing malformed-constraint diagnostic. Fresh current-head proof completed: 19 core native/AO rows PASS, plus 6 scope/isolation rows PASS (separate attempts, not 25 unique capabilities). Both attempts verified zero owned survivors.
A distinct current-head attempt stopped when the observer assumed one permission request per task: after the requested write was approved, Pool asked to read back its result. The native ledger retains both distinct requests. That attempt remains FAIL/incomplete. The corrected private observer explicitly rejects additional requests and keeps the protected-file, process and settled-turn gates unchanged. The later fresh attempt passed; its secondary-rejection branch was not needed in that reproduction. No product permission gate was weakened. Clean tested source/build: Core receipt SHA-256 Actual Electron/UI proof: BLOCKED capacity. The measured current lockfile forecast needs 2.43 GB including the 700 MiB free reserve. Even removing every remaining permitted regenerable Go-cache entry would provide only 1.71 GB. No npm ci or Electron launch was attempted; no new PNG exists. All 24 UI checkpoint rows remain in the private checkpoint/attachment queue. Published screenshot attachments are independently BLOCKED because no approved uploader is available. AO auth remains configured/unverified; no native safe status probe, login/logout operation, default/manual-mode run, provider-account maintenance UI, model-picker UI, actual UI reconnect/daemon replacement or negative-history restore is established by these functional tests. TUI, Auto, Reviewer and handoff remain intentionally excluded and truthfully refused. Existing initials branding remains the documented license fallback. This is supported functional proof with explicit qualification gaps, not full integration PASS. Original comment by @nikhilachale on 2026-10-21T08:26:09ZPool final-proof status — implementation pushed, functional proof scoped; full integration qualification remains incomplete. Current PR head and clean tested source: The surgical repair adds native Account setup/logout plans, keeps setup reachable for unknown auth, retains provider metadata without inventing groups from model IDs, explicitly refuses terminal refresh, reconciles latest main/API artifacts and the unmerged migration, and preserves existing schema-error precedence. Main's model visibility subsystem was reused. Excluded integrations and the readonly harness skill were not edited. Current-head native/AO proof is 19 core rows PASS + 6 scope/isolation rows PASS, with zero owned survivors. Detailed current-head evidence and original failure distinctions. Runtime AO SHA-256
Actual UI/attachments remain BLOCKED. The measured current lockfile forecast requires 2.43 GB including the 700 MiB reserve; only 1.71 GB would be available even after all remaining permitted Go-cache removal. No npm ci, Electron launch or capture was attempted. There is also no approved working attachment uploader. These are separate blockers; no fake image, URL or evidence commit substitutes for them.
Original failures remain separate: sparse Go workspace-module omission before generation/build; the first pushed head's SQLite regression, corrected by The hygiene scanner still flags the original PR's 12 evidence files, old host-path receipts and proof workflow. They are retained unchanged because no verified attachment relocation is available. No new evidence, reports, binaries, PNGs or workflow changes were committed. Current command/receipt/helper hash indexes, source/native binaries, isolated histories/workspaces, checkpoint queue and prior dirty repair checkout are retained privately. Public excerpts exclude credential-reader locations, private instructions and profiles; no uploaded executable reproduction bundle is claimed. Final owned cleanup: PASS; every leased child ended and per-attempt PID/kernel-birth checks reported zero survivors. Native installs/profiles/history, original evidence/comments and both source worktrees remain retained. Routine unit/race/lint/vet/typecheck/build-fixture/runner-fixture suites ran only in CI. Current source compile/API generation and actual live proof ran only under the shared lease. Linux functional proof does not qualify native macOS/Windows behavior. No full integration PASS: strict auth verification, actual UI journeys and published screenshot attachments remain unqualified. The next UI proof needs sufficient permitted capacity and an approved uploader; it does not need a new account or credential reset. |
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: true
|
🏆 Review leaderboardNov 5–12, 2026 · UTC
Ranked by distinct external PRs reviewed, then review rounds, then PR comments. Self-activity and bot activity are excluded. Show 6 more reviewers
|
Account migration verification
|
Resolve conflicts with main's Hermes harness, ACP transport, registry, SQLite and API changes, keeping main behavior and adding Pool alongside: - acp: keep InstallCheck/RetryStartupExit with ValidateSessionReady and adapt bindSessionValidation to main's (response, model) failure hook - acp tests: pass main's steering argument to conversationCapabilities - nativeacp: keep main's SessionMeta/QuickStop/steering fields, add ValidateSessionReady and Preflight without duplicating ValidateInitialize - harness, registries, DTO enums, OpenAPI and schema.ts: list hermes then pool - agentauth: keep both plans and give Pool its documented POOLSIDE_API_KEY provider-key method, required by main's account-setup coverage test - sqlite: reconcile and verify both Hermes and Pool; ledger Pool's migration in version order; baseline the Pool round-trip test on the migration immediately before it so it exercises only Pool's Up/Down - TaskComposer: route Pool through main's submitMode/launchFallback as a Chat-only agent - settings: keep main's showSetup account action, which covers Pool setup - workflows: keep main's manual-only harness artifact build Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pool's turn-settings validator returned a plain error, so main's SupportsPermissionMode treated every mode as supported and the Chat permission picker offered Auto, which launch then refused. Wrap ErrChatPermissionModeUnsupported and assert the driver's checker agrees with the native mode mapping. Drop the duplicate CHAT_PERMISSION_MODE_UNSUPPORTED case the merge left in the conversation error switch; main already maps it earlier. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The qualification receipts, reconciliation JSON, evidence reports and UI screenshots are run artifacts. They remain available at the contributor's pinned fork commit linked from the PR description; repository docs keep only the Pool setup guide. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Main has shipped migrations through 20261114072016988, so Pool's 20261021073338393 would apply out of order on every existing database. Take a fresh ID above main's latest, append it to the ledger, and baseline the Pool round-trip test on the migration immediately before it. Also cover the latest schema in the repair-anchor test and assert that ACP bindings without an explicit capability opt-out keep protocol-declared capabilities. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Main reconciliation and ao-agent-harness receipt (current head)Head: What changed on this branchThe 23 conflicts were resolved by merging main. History was not rewritten (fast-forward from Defects found and fixed:
Per-step receipt
Gates still open
|
nikhilachale
left a comment
There was a problem hiding this comment.
Review: changes requested (1 required change)
Reviewed head 4d24e176a0 against main (merge base 28bc83fd95). The Pool Chat integration is well contained. It is Chat-only at spawn (rejects TUI before allocation), Auto is refused through the driver permission checker, terminal launch/restore are hard-refused, discovery reuses the native ACP catalog with a config/credential fingerprint, and the reconcile path repairs a skipped migration. The shared ACP changes are safe for the other adapters: the new explicit-opt-out loop in conversationCapabilities only changes behavior for Pool, since piacp opts out of Approvals, which the protocol never sets. ValidateSessionReady is not bound on reconnect, and Launch/configure (which deletes the witness) only runs on a real launch, so a live host keeps its witness.
Required (inline):
- The Pool down migration can leave rows that break the restored CHECK, and it silently does nothing once a later harness sits before
fake. Mirror the guarded Hermes migration.
Optional suggestions (not required):
service/chat/service.goPreflightChat: the hardcoded Pool/Auto branch repeatschatAdmissionError(poolvalidateTurnSettingsalready returnsErrChatPermissionModeUnsupportedfor Auto throughSupportsPermissionMode). Dropping it would keep harness-specific policy in the driver.poolacp.validateSessionReadycomparesfilepath.Cleanpaths. If Pool ever reports a symlink-resolved cwd (for example macOS/private/varor a symlinked home), every completed turn would fail. Comparingfilepath.EvalSymlinksresults on both sides would be more robust.cli/doctor.gohas nopoolentry in its binary/version table, soao doctorwill not report Pool.adapters/agent/pool.ResolveBinarylistspool.exeeven though the integration refuses Windows. This is harmless, but it is inconsistent with the Unix-only scope.
The Pool down step narrowed the sessions CHECK even while pool rows existed, and matched only when Pool sat directly before 'fake', so the next harness migration would turn it into a silent no-op. Mirror the Hermes migration: refuse rollback while pool rows exist or the constraint shape is unknown, remove the Pool entry wherever it sits, and assert it is gone after RESET. The up step now widens any constraint ending in 'fake' and asserts Pool is present. Because Up now also widens historical variants, startup repair sets Pool aside before its exact historical replacements, as it does for Hermes, then appends it again. Tests cover refusal while a pool row exists, rollback after a later harness lands before 'fake', and the Up guard on an unknown constraint. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>


Code changes: +517 -21
Tests: +810 -4
Others: +823 -5
What
feat: integrate Pool native ACP Chat with scoped lifecycle proof
Original PR description, preserved
Binary assets in this source snapshot: 3 (excluded from line counts).
Latest repair and clean runtime-tested source:
a10f3551a0fc15f673a8026c65e461da2324f157. Final status, completion matrix and current-head CI · Detailed current native/AO evidence and retained failures. Exact-head CI: 25/25 successful. Actual Electron UI and screenshot publication remain blocked. Earlier source/CI/runtime snapshots below are retained as historical evidence and are superseded by this repair's linked status.Unix Chat only; TUI, Auto, reviewers, handoff and unqualified optional ACP features refused.
Current status
Current source snapshot:
56670397922d374a83c4344ba8999e47457828d7(basedb77f945db58ce1d9970e99e7e9b68396f38c78b). Open for review; GitHub mergeabilityMERGEABLE, merge stateBLOCKED. Exact-head CI: pending (3 successful, 20 queued/running, 0 unsuccessful among 23 reported checks). CI run 1 · CI run 2 · CI run 3 · CI run 4. Green source CI does not establish current native/UI runtime proof or full integration readiness.This fix created only clean isolated source worktrees and temporary review reports; it created no AO/native sessions, provider profiles or owned runtime processes. No local build, gofmt, test, fixture, provider or UI run was performed.
Historical runtime build: Historical AO/UI reconnect/capture source
fb60bd4097ebd4109bbe7ee68e908f722e6ee03f; earlier full lifecyclea7eadca3d2f4ccbd7dd376933fc8a57793430209. Released Pool1.0.16 executable SHA256db70340b41c8bd76dbe2b5926a6e46cda804789b9774b3d4cfba3591f0de02f2; final AO executable SHA256b30f2dfa7b1f1c796c5772cb96d7d5c32b4bfb53c39f2c4ff7ca669f2cca8ded. Attempt003 and the final reconnect are distinct. Report · evidence index, original receipts, image/provenance hashes. Receipt SHAs/bytes are unchanged. Linux/VPS proof is not Windows/macOS native proof. Runner SHA2568887aa2e7101b9e6d3950bfca08ffddf2e1e6c1c3f1d030c8dec2676443aecearemains historical where reported; manual/standalone proof is not a strict runner PASS.Changes since runtime proof: Source now discovers configured models through a short native ACP session, retaining provider-native model IDs/current default and a configuration-sensitive cache fingerprint. Settings has an actual manual setup plan and user guide, and unsupported OS is rejected before Chat startup. Explicit malformed provider configuration is rejected before native artifacts, but after AO allocation; this is not a pre-allocation malformed-config guarantee. TUI/Auto capability refusals remain separate. Setup documentation records the intentional initials avatar fallback; Poolside trademark guidance does not establish a verified logo grant for this avatar use. None of these changed paths has new native/UI runtime proof. The first source-CI attempt at cd4df74 reported goimports, preallocation and error-string lint issues; the follow-up corrects those source details. Original failed lint job is retained; no AO screenshot exists because this was a source-lint job.
The matrix separates current verification from recorded historical passes;
NOT_RUNdoes not erase those passes. Excluded interfaces/manual model input are deliberate supported-scope decisions.Historical implementation and evidence
Adds nativeacp agent/Chat registration, project/task defaults, pre-allocation TUI/Auto rejection, product labels, DTO/generated contracts and migration0208 with historical/burned-schema reconciliation. Native ACP identity pool-acp1.0.16 is checked. Optional images/embedded context/resource links/elicitation/nested agents/terminal output remain scoped off. Private bounded unnamed SessionStart hooks preserve native/user defaults and tools; a matcher is required. Hooks can initialize lazily and fail open, so AO checks a session/workspace witness after prompt completion; this is an observational check, not a pre-tool security boundary. Detached adoption retains original context. The Kill API exposes no freed count. Historical fb60bd4 source CI was 23/23 green; evidence d4ccc7f was 5/5/150 fixtures. Z.ai404, missing matcher, eager restored-context check, post-reject model refusal, observer assertion failures and resource-stop build remain distinct.
historical receipt 1 · historical receipt 2
Reproduction and retained-original limitations
The pinned reports/JSON retain action/result, file/token effects, process identity, cancellation, native-history continuity and capture assertions. A complete sanitized executable reproduction bundle is not established for every manual helper: exact helper bytes/hash and invocation, applicable local-contract bytes/hash with probe argv and interrupt/readiness/draft-cleanup settings, and complete native/AO fingerprints are unavailable where not recorded in the linked receipts; recoverability of those original inputs is unknown. No missing command, contract, hash or result is invented. Later reproductions must use a new attempt and cannot replace original failed/missing evidence.
Recorded owned-process/session/daemon/Electron cleanup is historical. Source worktrees, native profiles/history, dirty scratch workspaces, binaries and raw logs were reported retained privately; a complete current inventory/location/hash and recoverability of every original class are unavailable in the public record. Current recoverability is unknown; this reporting fix reconstructs no missing input and silently substitutes none. Share sanitized artifacts only; exclude credential stores, raw DBs and provider-profile directories. Preserve the retained failures and original screenshots.
Why
The contributor is moving this harness submission to @milesatao-svg. The linked source preserves the implementation and review history.
Before / After
a10f3551a0fc15f673a8026c65e461da2324f157a10f3551a0fc15f673a8026c65e461da2324f157Original product behavior and existing UI evidence are included in the preserved description and copied evidence comments.
How
Copy the source head into
milesatao-svg/agent-orchestrator:codex/copy-pr-6551-pooland target upstreammain.Testing
Historical test results and screenshots are reproduced in the evidence comments with their original sources. Fresh validation is tracked by this PR's GitHub checks.
Checklist
Account migration verification
a10f3551a0fc15f673a8026c65e461da2324f157.milesatao-svg/agent-orchestrator:codex/copy-pr-6551-pool. Base:main.