Skip to content

feat: integrate native OpenJarvis code-assistant TUI - #8205

Draft
milesatao-svg wants to merge 7 commits into
OrchestratorInc:mainfrom
milesatao-svg:codex/copy-pr-6537-openjarvis
Draft

milesatao-svg wants to merge 7 commits into
OrchestratorInc:mainfrom
milesatao-svg:codex/copy-pr-6537-openjarvis

Conversation

@milesatao-svg

@milesatao-svg milesatao-svg commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Current head, 2026-11-14 IST: 5e109bdf47 (native-review fixes for the input witness, failure marker, withheld-key auth and the unused permission sentinel; all 4 threads resolved), with main adf9f6a303 merged in (no history rewrite). GitHub reports MERGEABLE. Draft, not merge-ready. Receipt and change list.

Supported scope: the native TUI only. There is no native session ID or restore (typed AGENT_RESTORE_UNSUPPORTED), and no Chat, switching, handoff or reviewer. Input is one plain line; every AO permission mode runs with native security.

Open checks:

  • CI on this head: all 4 workflows are action_required and need maintainer approval for fork runs
  • Live TUI audit (ao-agent-harness step 4). Expected unsupported results: tui_native_id, tui_restore
  • UI audit with screenshots (step 5)
  • Reviewer acceptance of the activity-dispatch touchpoint as not applicable (no native hooks; terminal detection)
  • Maintainer decision: keep, scope down or revert the shared conpty/surface.go resize change. It affects every TUI harness on every OS and needs cross-harness proof (Codex and Claude Code inline frames after a shrink, at least on Windows)

Research reports and screenshots were removed from the diff; the historical links below still resolve to their original commits.

Code changes: +962 -13
Tests: +564 -15
Others: +563 -4

Account migration from @nikhilachale to @milesatao-svg. This draft continues original PR #6537 at original head 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac. Original commits and commit authorship are preserved. Descriptions and evidence below include historical results with their original sources.

What

feat: integrate native OpenJarvis code-assistant TUI

Original PR description, preserved

Rebase follow-up, 2026-10-22: Current head 1a9055a5d, rebased as a consolidated integration onto main c3d22031e. Retains Hermes registration and main terminal-admission protections, regenerates API types for both harnesses, separates daemon test helpers, and covers 39 installer harnesses. The unmerged OpenJarvis migration uses timestamp 20261022042122070; upgrade/downgrade tests retain a Hermes session. CI follow-ups fix an inherited duration-variable lint name, keep the cloud heartbeat fixture line unfinished, and align the update-notice test with its existing safe View details fallback after an installer query error (no production UI change). Backend build/vet, Linux-target lint (zero findings on an LF-normalized temporary copy), focused admission/ConPTY/auth/installer/migration/setup-timeout checks, API drift, both frontend typechecks, 8 avatar tests, all 54 update-notice tests, and all 161 Product UI tests pass locally. Full Windows suites remain non-green: renderer rerun has 492 passing and 24 failing files (8667 passed, 213 failed, 8 skipped tests), including packaging/path/runtime/symlink cases; backend also has platform/environment failures. Linux/race, native macOS, Docker and browser gates require CI. At d32353aa2, 23 of 24 CI checks passed; its sole failure was the now-corrected notice assertion. Current-head CI is complete: all 24 checks passed at 1a9055a5d (checks); GitHub reports MERGEABLE. Historical native/UI qualification limits remain unchanged.

Lint follow-up, 2026-10-22: 9145086c3 fixes the two Linux CI lint findings in openjarvis/auth.go: use http.NoBody for GET requests and explicitly discard the response-body close error. Auth-focused tests pass locally. Full Windows backend tests and lint are not green (platform/PATH/symlink and CRLF findings); the existing OpenJarvis workspace-path assertion also fails on the unchanged PR head. Linux/race suites are unavailable locally (no WSL/Docker or CGO). Those conflicts were resolved by the subsequent rebase described above. Historical qualification limits below remain in effect.

Binary assets: 4 (excluded from line counts).

Current publication, 2026-10-15: 6a7edfa2a0b0ca1918446597de78fdb4584c68aa. Incomplete qualification; AO UI evidence BLOCKED; not merge-ready. Main conflicts reconciled. Exact-head CI is recorded below; no local routine suites.

Published typed restore/input/permission refusal handling and current-main reconciliation. Historical runtime77574094 demonstrated file/follow-up effects, child cancellation and fresh Kill/no-late/control isolation. Independent project instructions failed; post-cancel send was unconfirmed and was never retried. History is in memory: native ID/exact restore/Chat remain unsupported; no replacement identity was invented.

Current report and scoped evidence. Historical runtime/source attribution is retained there.

Exact-head CI (2026-10-15T07:12:44.140233+00:00): complete at 6a7edfa2a0b0ca1918446597de78fdb4584c68aa — 25 SUCCESS (25 reported; 25 expected including the merged Go aggregate gate). Current checks. Mergeability MERGEABLE, merge state BLOCKED; draft handoff. CI does not qualify missing native/UI/provider/OS gates. Canonical main subsequently advanced to48c4ef5e with only two non-overlapping preview-process files; GitHub reports this head mergeable. These completed checks belong to the stated publication head, not a newly rebuilt merge with that later base.

Preserved earlier description (historical status, validation and evidence snapshots)

Binary assets in this source snapshot: 4 (excluded from line counts).

Official native code-assistant orchestrator TUI; scoped plain single-line input. In-memory history has no native ID/restore; restore, Chat, switch and handoff refused.

Current status

Current source snapshot: 77574094445328af21209f6d32bde40a6cb03143 (base db77f945db58ce1d9970e99e7e9b68396f38c78b). Open for review; GitHub mergeability MERGEABLE, merge state BLOCKED. Exact-head CI: pending (12 successful, 11 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 source a4bc98eb0a759b4bcf38cf7a6481c45a9ac77315; official OpenJarvis CLI1.0.4 fixed upstream source 0eb7811df96cf139e6bd6d7fcc9b5cc4ff1418ff. Final008 and earlier004/006/007 are distinct; native binary fingerprint is not supplied in the reviewed relocated manifest. Report · evidence index, original receipts, image/provenance hashes. Receipt SHAs/bytes are unchanged. Linux/VPS proof is not Windows/macOS native proof. Runner SHA256 8887aa2e7101b9e6d3950bfca08ffddf2e1e6c1c3f1d030c8dec2676443aecea remains historical where reported; manual/standalone proof is not a strict runner PASS.

Changes since runtime proof: Source maps the unmodified official OpenJarvis circular PNG to the avatar, with pinned upstream source, Apache-2.0 license and authors notices in the sidecar. Actual current-avatar rendering and native/UI journeys are NOT_RUN. Current-main reconciliation preserves the integration and upstream migration ledger; current source/CI state is recorded above.

The matrix separates current verification from recorded historical passes; NOT_RUN does not erase those passes. Excluded interfaces/manual model input are deliberate supported-scope decisions.

Area Supported scope Verification Proof or reason
Registration and UI Official native code-assistant orchestrator TUI; scoped plain single-line input. In-memory history has no native ID/restore; restore, Chat, switch and handoff refused. NOT_RUN Adapter/registry/API/generated/product paths are present; current Settings/task journeys not reverified.
Models Native model IDs/manual input; discovery failed in the historical diagnostic audit. No strict catalog pass claimed. NOT_RUN Historical native selection separate from current picker/default verification.
Setup Native official install/manual prerequisites; Settings and guide path need current-source review. NOT_RUN Installer metadata/API enum alone is not onboarding proof.
Readiness Scoped native CLI/config prerequisites and unsupported surfaces; current-source OS/setup checks pending. NOT_RUN Historical native proof does not qualify all advertised OS/current config cases.
Identity and logo Official unmodified OpenJarvis PNG; Apache-2.0/authors sidecar and avatar mapping. NOT_RUN Pinned source/license sidecar; source disposition done, actual current avatar UI NOT_RUN.
TUI and Chat TUI supported; Chat/restore/handoff excluded for native capability reasons. Historical AO screenshots retained. NOT_RUN Historical actual AO images; excluded interfaces are deliberate.
Permissions and instructions Additive native config preserves safety/approvals; confirmation blocks ordinary sends, generic HTTP500 UX remains documented. NOT_RUN Historical scoped PASS/limits; current live path NOT_RUN.
Lifecycle Historical executing Ctrl+C and independent Kill/control; no native restore. Earlier late artifact/wrong project failures retained. NOT_RUN Historical scoped PASS/limits; current live path NOT_RUN.
Evidence Pinned historical source/native/runner/UI receipts, attempts, images and provenance; missing originals disclosed. NOT_RUN Historical evidence retained; current native/UI proof NOT_RUN. Reproduction-input limitations below remain explicit.
CI and handoff Exact source checks/merge snapshot; incomplete live qualification and historical cleanup. NOT_RUN Exact-head CI pending (12 successful, 11 queued/running, 0 unsuccessful among 23 reported checks); mergeability MERGEABLE. Source CI scope only; current native/UI qualification NOT_RUN.

Historical implementation and evidence

Official native jarvis chat --agent orchestrator preserves SystemPromptBuilder/native safety through additive private configuration. Input is plain single-line, at most3000 UTF-8 bytes; control characters, slash commands, ambiguous prompt markers and uncertain pending delivery are refused. Native confirmation remains intact and blocks ordinary sends; generic HTTP500/refusal UX with request IDs is a known limitation. AO observation cadence is30s, separate from native settlement. Native shell AO routing persisted a session checkpoint. Native Rust rate-limiter warning/model-discovery/install limitations remain documented; no strict runner PASS is claimed. Historical source a4bc98e had 23/23 checks green after unchanged-head rerun of TestKillReq. Native001–003/AO004 routing-resize,006 later approved rerun/late artifact and007 wrong configured project failures remain separate from final008.

historical receipt 1

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

Before After
Submission in PR #6537 from @nikhilachale at 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac Submission from @milesatao-svg at the same commit 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac

Original 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-6537-openjarvis and target upstream main.

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

  • Original head SHA and commit authorship preserved.
  • Original description and evidence provenance included.
  • Fresh CI validation on this submission.
  • Resolve current main merge conflicts.

Account migration verification

  • Exact source head copied: 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac.
  • New fork branch: milesatao-svg/agent-orchestrator:codex/copy-pr-6537-openjarvis. Base: main.
  • Evidence comments retain their original author, timestamp, source URL and result wording.
  • Fresh validation status is tracked by this PR's GitHub checks.
  • Source mergeability at migration: conflicts with current main; reconciliation pending.

@milesatao-svg

Copy link
Copy Markdown
Contributor Author

Copied source evidence 1/1

Historical material from original PR #6537, copied for this account migration. Source PR head when copied: 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac. Each original comment below retains its own author, timestamp and source link. Its test results and screenshots describe the original run and commit. Fresh checks for this submission are shown on this PR.

Original comment by @nikhilachale on 2026-10-11T08:32:48Z

Original comment

Final source a4bc98eb0a759b4bcf38cf7a6481c45a9ac77315: all 23 CI checks successful. Actual scoped proof report includes retained failures, exact serving-binary/generation provenance and cleanup.

Actual AO Electron openjarvis

Actual AO Electron openjarvis

These are real original VPS AO Electron captures. Unsupported native surfaces remain explicit; no strict runner PASS or unsupported history restore is claimed.


Original comment by @nikhilachale on 2026-10-12T16:52:00Z

Original comment

Harness-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.

Recorded harness evidence — see scope above


Original comment by @nikhilachale on 2026-10-12T18:08:10Z

Original comment

Canonical main conflicts reconciled at d45228394ed6, preserving previous source/evidence ancestry and every existing harness report/screenshot byte. Reconciliation receipt.

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 2026-10-12T18:17:19Z

Original comment

Canonical main conflicts reconciled at 500ae68dc09a, preserving previous source/evidence ancestry and every existing harness report/screenshot byte. Reconciliation receipt.

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 2026-10-12T20:58:49Z

Original comment

Canonical main conflicts reconciled at 473ed769bfa8, preserving previous source/evidence ancestry and every existing harness report/screenshot byte. Reconciliation receipt.

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 2026-10-15T06:35:49Z

Original comment

Published, incomplete qualification — AO UI evidence BLOCKED; not merge-ready.

Head 6a7edfa2a0b0ca1918446597de78fdb4584c68aa; main 54b28374912cb2771f17b8cc32a9482e6d607bce reconciled without force-push. Published typed restore/input/permission refusal handling and current-main reconciliation. Historical runtime77574094 demonstrated file/follow-up effects, child cancellation and fresh Kill/no-late/control isolation. Independent project instructions failed; post-cancel send was unconfirmed and was never retried. History is in memory: native ID/exact restore/Chat remain unsupported; no replacement identity was invented.

Publication report · All ordered screenshot checkpoints

Every applicable latest UI checkpoint remains NOT_RUN because this stage had no genuine isolated Electron dependencies and insufficient disk preserving512MiB. API/native receipts are auxiliary, never screenshot substitutes. No image was fabricated.

Exact publication-head CI is being checked separately below; old green heads do not qualify edited/merged source. No routine VPS tests, lint, typecheck, gofmt or CI emulator ran. Linux/provider evidence does not establish other OS/provider coverage. Original failures/profiles/logs remain intact privately. No skill6509 writes, production changes, merge, release or deployment.

Exact-head CI (2026-10-15T07:12:44.140233+00:00): complete at 6a7edfa2a0b0ca1918446597de78fdb4584c68aa — 25 SUCCESS (25 reported; 25 expected including the merged Go aggregate gate). Current checks. Mergeability MERGEABLE, merge state BLOCKED; draft handoff. CI does not qualify missing native/UI/provider/OS gates. Canonical main subsequently advanced to48c4ef5e with only two non-overlapping preview-process files; GitHub reports this head mergeable. These completed checks belong to the stated publication head, not a newly rebuilt merge with that later base.


Original comment by @Meraj-08 on 2026-10-20T09:31:04Z

Original comment

Screenshots

Install button-
s07b-oj

Health-
s11-oj-health

chat-
s18b-oj

Spawn Workers -

image

Original comment by @nikhilachale on 2026-10-21T07:14:19Z

Original comment

Fresh testing-only retest starting for openjarvis at current PR head 6e86bca3b1077b201df06aa9df0c269d4cdb2089 using harness skill bfb28b79f90df0a56892e8d8b1d3139dea1148db. One bounded serial attempt (25-minute maximum; eight-minute output-stall watchdog); isolated source/runtime. Previous evidence is historical and will not count as fresh proof. Full 12-area matrix and screenshot checkpoints will follow in a new comment. No implementation, pushes, production restart, or excluded PR changes.


Original comment by @nikhilachale on 2026-10-21T07:26:48Z

Original comment

Bounded TESTING-ONLY retest: BLOCKED — integration incomplete. AO runtime NOT_RUN; strict runner FAIL at infrastructure startup. AO UI attachment evidence: BLOCKED.

Tested source/frontend SHA: 6e86bca3b1077b201df06aa9df0c269d4cdb2089. Staged skill SHA: bfb28b79f90df0a56892e8d8b1d3139dea1148db; all 15 staged files byte-match that commit. Fresh GitHub head recheck at 2026-10-21T07:25:48.667611+00:00: 6e86bca3b1077b201df06aa9df0c269d4cdb2089. Detached source remains clean; no source/branch changes, push, merge, review publication, install/upgrade, login/reset or old-comment edits were performed.

Recorded comparison main SHA: 474962e31bea34b96588e8d91ea7a46d82557f8c (not merged). Local comparison is 14 main-only / 29 tested-head-only commits. Current GitHub base is 8fbec789517cb2b4ae979f81bde63f8a3242f643; GitHub reports mergeable=true, mergeable_state=blocked. This is not a merge-readiness claim.

Build/native provenance and first failures. Linux amd64; Node v24.21.0, npm 11.19.0, Go go1.27.1. Fresh native jarvis --version reports jarvis, version 1.0.4; actual launcher SHA-256 f2d2ba4d6d928b4f038caa9d0612d75095f6c6e9ed1dfc6c65c803a5348cd0b0. Installed native source fingerprints are retained privately; zero installation files changed since worker start. No AO binary/version/hash exists from this attempt. No serving frontend/daemon, AO session/native-session digest or terminal generation was created.

Native safe probes on 2026-10-21 UTC (PASS means command completion, not provider/auth conformance):

Command UTC interval Exit Output SHA-256
jarvis --version 07:17:47–07:17:49 0 24641bf74b24255d1458bafc5aefa6264b42438630165eb2355d383677f8381f
jarvis --help 07:17:49–07:17:50 0 0b9b1fdc33d95bf6e16cb49aee86db9cbeb92a5eadc39425ae9eca65a5321669
jarvis chat --help 07:17:50–07:17:51 0 c5a819b3ab0beed5706b9567a105b1df326a31f49af431f278b63746304d6a28
jarvis ask --help 07:17:51–07:17:52 0 530b1cdc2da68b22e5005876a7478c14caf4e587a81a813707712d0a6a1b310e
jarvis model list --help 07:17:52–07:17:53 0 d615991c824916bbcd2de163e0197eae2ba36840f4e2155ce8a02d3544f00a57
jarvis model list 07:17:53–07:17:55 0 c8da530ed6d45ec6fdc1bb0fd7166c4ecfbcecd1ad92188f48ddbcdd3f7f18bf

The model table contains genuine cloud row IDs gpt-4o, gpt-4o-mini, gpt-5, gpt-5.4, gpt-5-mini, o3-mini. These are native static suggestions from the configured client, not provider entitlements. Selected glm-5.3-flash is a free-form override and is absent from that suggested catalog. Local engines' rows were excluded, not misrepresented as cloud models/providers. Current-head modelcatalog includes a cloud Rich-table parser, contradicting the older prose's no-catalog claim; no product fix was attempted.

Distinct real attempts; later observations do not erase first failures:

Attempt/scope Command/result and time Classification
Minimal AO build go build -ldflags='-s -w' -o <runtime>/bin/ao ./cmd/ao, 07:18:25–07:18:45 UTC; exit -15 from owned disk guard; no binary. Existing Go caches, GOMAXPROCS=1, GOFLAGS=-p=1, GOMEMLIMIT=384MiB. BLOCKED infrastructure; first build failure preserved.
Native ask witness 001 jarvis ask --engine cloud --model glm-5.3-flash --agent orchestrator --tools file_write <unique-owned-file-task>, 07:19:28–07:19:29 UTC; exit 1, no file. Output SHA-256 86c09237afc22f18f48662d0547f4f0f4477e5603d66b18cd320f77622d54ade. FAIL before provider: engine selector says cloud unavailable for configured model. Not an authentication failure.
strict-001 Staged audit_ao.mjs with explicit tested source, actual intended canonical <runtime>/bin/ao, derived contract and new report basename; 07:20:29.059–07:20:29.177 UTC; exit 1; finalized strict report summary.status=FAIL, startup exit -2, results=[], no agents discovered. Strict FAIL, missing binary after disk-limited build. No provider/auth/TUI/lifecycle gates ran inside AO.
Native chat 002, focused independent diagnostic jarvis chat --engine cloud --model glm-5.3-flash --agent orchestrator --tools file_write, 07:21:08.765–07:22:38.798 UTC; actual empty composer observed 07:21:09.863 before /model and before one task. Native /model: cloud / glm-5.3-flash. Provider returned HTTP 429, code 1310, “Weekly/Monthly Limit Exhausted.” No file. Bounded 90-second stop, exit -15; PID 1492366 / kernel-start 161302158 confirmed gone. Output SHA-256 253fa390b3d17e596c3978a8e572adf7a48b6269f234b6180087497f84d667f7. Functional FAIL / provider prerequisite BLOCKED by quota. Credential presence and quota response are not verified authorization. Native REPL evidence only; no AO diagnostic continuation.

No further retries or AO diagnostic were attempted: there is no current AO binary and the native workspace-proof prerequisite failed. Native startup also warned that optional openjarvis_rust was absent; its “Rust extension building” banner reads default pending state and does not establish an actual build. No native installer/build was invoked.

Capacity and actual UI. Initial free disk 834 MiB; setup measurement 822 MiB, required floor 700 MiB. Pinned Electron 44.7.0 archive alone is 122,998,223 bytes (117.3 MiB), before extraction and 1,497 pinned packages. Dependencies absent. Serial Go build guard fired near the floor; observed free space 710 MiB, with another observation 708 MiB while teardown completed. Only 56 MiB of this worker's abandoned build temp was removed after exact-process checks; no shared cleanup. Post-cleanup free space about 763 MiB. RAM total 3,906 MiB; initial available 2,535 MiB, later 2,427 MiB. npm ci/Electron were not started because setup could not preserve the disk floor. Node heap budget was 384 MiB. No old frontend, symlinked dependencies, terminal playback or source pictures substitute for AO screenshots.

Reproduction/current-status receipt. Existing matching credential was resolved from the known authorized native store in memory, passed only in the child environment to isolated native probes (and prepared for the owned daemon), never argv/config/logs/public receipts. Original accepted profiles and native installation stayed unchanged. The independently derived contract SHA-256 is 460c1242e24e4dde9261b593bcfd3a3f1c892a80db2b33e4f6aac3014076affd. Sanitized contract shape (only actual local binary path replaced):

{"schemaVersion":1,"agents":{"openjarvis":{"binary":"<native-bin>/jarvis","versionArgs":["--version"],"integration":{"args":["chat","--help"],"expect":"--agent"},"session":{"args":["ask","--engine","cloud","--model","glm-5.3-flash","--agent","orchestrator","--tools","file_write","{prompt}"],"proofFile":"LOCAL_AUDIT_PROOF.json"}}}}

No fabricated interrupt or restoredReady contract: process-local native history has no exact-load command, and Ctrl-C catching alone does not establish executing-tool child reaping. Native model table was independently parsed; the generic runner's text-line mode cannot truthfully treat Rich headers/borders as model IDs, so its model-command field was omitted. These strict gates remain NOT_RUN, not PASS.

# In the pinned tested backend, with existing caches and isolated runtime output:
GOMAXPROCS=1 GOFLAGS=-p=1 GOMEMLIMIT=384MiB go build -ldflags='-s -w' -o <runtime>/bin/ao ./cmd/ao
# Use a fresh report basename and runtime-local TMPDIR; credential is env-only.
node <staged-skill>/scripts/audit_ao.mjs --agent openjarvis --repo <tested-source> --ao <runtime>/bin/ao --local-contract <runtime>/contract-current.json --report <runtime>/attempts/strict-001/report --startup-timeout-seconds 15 --timeout-seconds 30

The build invocation is shown for reproduction; this attempt was disk-guard terminated and produced no binary. Reusing an old binary/report would not qualify this SHA.

Completion matrix. Source-only observations are labelled; none fill live/UI gaps.

Area Supported scope Verification Proof or reason
Registration and UI TUI worker registry, API/generated types, shared task/settings UI BLOCKED Worker registration, generated enums and product label traced at tested head; no current-head daemon/Electron session. Source inspection is not runtime proof.
Models Native cloud suggestions plus free-form --model; shared Models/New Task controls BLOCKED Fresh native model list exit 0: six cloud IDs. Current source has a Rich-table parser, despite prose saying no catalog. AO refresh, both pickers, off/on/bulk persistence and actual picker effects NOT_RUN.
Providers Existing OpenAI-compatible Z.ai coding route; native multi-provider env configuration BLOCKED Only matching existing key loaded in child memory. Native REPL selected cloud/glm-5.3-flash but provider returned 429/1310 quota exhausted. No successful turn, verified auth, AO provider/account metadata/management or routing proof.
Setup Pinned official UV install plan; instructions-only AO configuration plan NOT_RUN Already-installed jarvis 1.0.4 observed. Install/upgrade/login/logout/account reset/change branches unauthorized. AO-page setup/error/retry/readback unavailable; manual guidance is not in-page onboarding qualification.
Readiness Native executable/config/model/composer; fresh AO launch/auth readiness BLOCKED Fresh version/help/model commands pass. Native ask rejects configured model before provider. Native chat reaches actual composer but hits provider quota. AO probe/readiness unavailable; configured-only never upgraded to authorized.
Identity and logo OpenJarvis label and shipped licensed avatar BLOCKED Product identity, AgentAvatar mapping and Apache-2.0 license inspected; native version positively observed. Actual AO row/avatar never rendered.
TUI and Chat Native terminal REPL; AO Chat separately gated BLOCKED Native terminal startup and /model observed; no completed file turn. AO TUI NOT_RUN. No native ACP entry or OpenJarvis Chat registry entry found. Chat API rejection and all shared Chat UI subjourneys NOT_RUN; no N/A runtime claim.
Reviewer Independent reviewer suitability/registration/picker/run/result/cancel/rerun NOT_RUN OpenJarvis absent from reviewer vocabulary/registry; native prompt-driven review suitability remains unqualified. No AO reviewer run/target/head/result. External GitHub reviewer publication expressly unauthorized; absence of wiring is not native N/A.
Permissions and instructions Native security plus private instruction overlay; actual effective modes NOT_RUN Current adapter source accepts AO mode labels using native defaults; prose claims default-only. No live permission/approval/rejection or ordinary-task private/project instruction consumption established. Wire roles unknown. No bypass/downgrade performed.
Lifecycle Follow-up, active-tool cancel, Kill/archive, exact restore, Kanban return, reconnect/recovery NOT_RUN No AO session created. Native history is process-local and adapter refuses restore in source; actual AO refusal untested. No executing-command PID/start/delayed-effect cancellation witness. Native SIGTERM cleanup is teardown, not cancellation conformance.
Evidence Current source/native/skill fingerprints, distinct attempts, actual AO images/attachments BLOCKED Strict failure and native failures preserved; 24 checkpoint rows below. Zero AO PNGs and zero attachments. No approved attachment uploader: full integration PASS unavailable.
CI and handoff Current-head checks, divergence/conflict reporting and clean owned teardown BLOCKED 25 reported check runs succeed at tested SHA; no failing/in-progress check runs. Legacy combined status pending with zero status contexts. GitHub mergeable=true, mergeable_state=blocked. Runtime/UI gaps remain; PR-description evidence link pending parent update.

Complete screenshot checkpoint index. Created before any UI action. Every row has screenshot unavailable / attachment BLOCKED: actual Electron capacity blocked and no approved GitHub attachment uploader. Zero captures exist; the private attachment queue is empty. No public screenshot checkpoint passes. Pre-session page identities, capture timestamps, serving binary hashes, session/generation and image hashes are unavailable because those screens were never run, not reconstructed after cleanup.

Checkpoint Action → observed result Functional status Capture/attachment status and reason
harness-listed Open Agent Harness row → no actual AO UI BLOCKED BLOCKED / unavailable. No current-head daemon or Electron runtime; registry/API/UI wiring inspected only.
install-offer Show Install method/offer → no actual AO UI BLOCKED BLOCKED / unavailable. Pinned UV install plan inspected; no real AO page.
install-run Install progress → result → no actual AO UI NOT_RUN BLOCKED / unavailable. Provider/native-agent installation or upgrades explicitly unauthorized.
auth-before Show existing credential state → no actual AO UI BLOCKED BLOCKED / unavailable. Matching saved key resolved in memory; fresh AO auth unavailable, never verified from presence.
auth-action AO Login/Setup → return → no actual AO UI NOT_RUN BLOCKED / unavailable. Native/account login/reset not authorized; AO instructions-only setup plan inspected.
auth-ready Fresh Agent Harness auth result → no actual AO UI BLOCKED BLOCKED / unavailable. No AO readiness probe; independent native receipt is a separate scope.
auth-maintenance Account/provider change/logout/retry → no actual AO UI NOT_RUN BLOCKED / unavailable. Account mutation/reset prohibited; no accepted profile writes.
harness-models Models listing/refresh loading/result → no actual AO UI BLOCKED BLOCKED / unavailable. Fresh native Rich table cloud rows observed; AO catalog/parser inspected but no daemon/page.
model-visibility Off/on/bulk → save/reload/picker effects → no actual AO UI BLOCKED BLOCKED / unavailable. Shared disabledModels control documented; no isolated AO settings/UI runtime.
provider-management Provider metadata/configuration → effective routing → no actual AO UI BLOCKED BLOCKED / unavailable. Existing matching route loaded in child environment; AO management/routing not exercised.
task-agent New task → OpenJarvis selected → no actual AO UI BLOCKED BLOCKED / unavailable. Worker registry/product label present; UI not provisioned.
task-model Model picker → effective selection → no actual AO UI BLOCKED BLOCKED / unavailable. Native free-form override and static cloud rows observed separately; AO picker unavailable.
task-running Submit → Kanban card/native output/result → no actual AO UI BLOCKED BLOCKED / unavailable. No current-head AO daemon/session. Standalone native proof cannot fill this row.
task-followup Actual composer follow-up → response → no actual AO UI NOT_RUN BLOCKED / unavailable. No owned AO task/controller.
task-cancel Executing-tool Stop → settled/reaped → no actual AO UI NOT_RUN BLOCKED / unavailable. No AO session or active-command PID/kernel-start/delayed-side-effect witness.
kill-action Archive control/confirmation → terminated → no actual AO UI NOT_RUN BLOCKED / unavailable. No owned AO session; no Kill acknowledgement claimed.
archived Same terminated entry in archive → no actual AO UI NOT_RUN BLOCKED / unavailable. No AO session created.
restore-action Restore → same native ID/ready → no actual AO UI NOT_RUN BLOCKED / unavailable. Native REPL history is process-local; adapter returns ErrAgentRestoreUnsupported. Live AO refusal untested.
kanban-return Same restored card → continuation → no actual AO UI NOT_RUN BLOCKED / unavailable. Exact restore unavailable; no live AO refusal or board exercised.
chat Chat New Task/composer/stream/output/history/approve/reject/Stop/reconnect/recovery → no actual AO UI NOT_RUN BLOCKED / unavailable. No OpenJarvis ACP driver in registry/native CLI. Native command named chat is a terminal REPL, not AO Chat; actual API rejection untested.
reviewer-select AO reviewer picker/target/head/trigger → no actual AO UI NOT_RUN BLOCKED / unavailable. OpenJarvis absent from reviewer vocabulary/registry. Ordinary worker review is not qualification; live refusal untested.
reviewer-running Reviewer run/output/cancel/retry → no actual AO UI NOT_RUN BLOCKED / unavailable. No registered OpenJarvis reviewer; no reviewer execution/publication authorized.
reviewer-result Reviewer result/verdict/rerun/head → no actual AO UI NOT_RUN BLOCKED / unavailable. No AO reviewer run. GitHub review publication expressly unauthorized.
reconnect-cleanup Reconnect/recovery/cleanup before→after → no actual AO UI NOT_RUN BLOCKED / unavailable. No AO UI/controller existed. Exact owned-process shutdown verified separately at handoff.

Chat subjourneys explicitly NOT_RUN: new Chat task, shared composer/follow-up and streaming, output/history/native load, approve, reject, witnessed active-tool Stop, UI reconnect, daemon/host recovery, Kill/restore/new instructions, negative capability/restore controls. No OpenJarvis ACP driver is registered; native jarvis chat is a terminal REPL. No live AO rejection was tested, so this is not qualified N/A. Reviewer select/target/head, run/cancel/retry and result/rerun are NOT_RUN; no reviewer registration and no external reviewer publication permission. Install/login/provider-change branches remain NOT_RUN under the testing-only authorization; unsupported-source observations do not count as UI refusal proof.

Current-head CI only. All routine unit/race/lint/vet/typecheck/build-fixture/runner-fixture suites were CI ONLY; none were run locally. Live minimal backend build and native commands above are separate. Fresh checks read at 07:21:35 UTC: 25/25 check runs completed/success at the tested SHA; zero failing or pending check runs. Legacy combined-status API returns pending with zero status contexts; no missing context/job is invented. These checks do not prove authenticated provider, AO runtime or current-head Electron journeys. Reported jobs:

Job Current-head result
build-test completed / success
sqlite-race (0) completed / success
chat-race (0) completed / success
sqlite-race (1) completed / success
session-store-race completed / success
sqlite-race (3) completed / success
windows-workspace completed / success
api-drift completed / success
chat-race (1) completed / success
cloud-build-test completed / success
sqlite-race (2) completed / success
sqlc-drift completed / success
backend-core (1) completed / success
backend-core (0) completed / success
lint completed / success
renderer-smoke completed / success
container completed / success
native (windows-latest) completed / success
native (macos-latest) completed / success
test completed / success
native (ubuntu-latest) completed / success
mac-update-helper completed / success
scan completed / success
live-feeds completed / success
linux-proof-artifact completed / success

Cleanup and next action. Cleanup receipt at 2026-10-21T07:24:16.818214+00:00: PASS for owned teardown. All 10 command PID/kernel-start identities checked; zero same-identity survivors or owned process-group descendants. Strict runner finalized with daemonStarted=false / stopped=true; its selected loopback port refuses connection. No AO session, Electron, production restart or global stop existed. Both native witness files remain absent; this is failed task proof, not a delayed-command cancellation test. Git source remains clean, including untracked files; native installation modified-file count is zero. Retained operational evidence stays exclusively outside Git in this PR's sibling runtime: sanitized logs/receipts, native-source fingerprints, derived contract, isolated noncredential config/state and empty workspaces, strict report/issue files, full checkpoint index and empty attachment queue. No evidence/PNG committed or uploaded elsewhere.

Unresolved: provision enough disk while preserving the floor; resolve provider quota through the account owner without resetting login; qualify the configured custom model route separately (ask currently rejects it); rebuild this exact head; rerun the strict audit with a fresh basename; independently verify actual AO readiness/models/providers/permissions/TUI/refusals and all supported UI/Chat/reviewer paths. Source-only docs/catalog/permission discrepancies remain for authorized product work, not this worker. The parent must attach any future safe actual AO images through an approved route; AO UI attachment evidence stays BLOCKED. PR-description link to this new comment is pending the parent; description was not changed. No full integration PASS is claimed.


Original comment by @illegalcall on 2026-10-23T18:37:06Z

Original comment

Independent runtime validation at 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac (macOS arm64, isolated lab)

I'm an external validator, not the implementation owner. I tested the current force-pushed head 1a9055a, which is rebased onto main c3d22031. The PR body's "current publication" 6a7edfa and the earlier 77574094/a4bc98eb/6e86bca runtime receipts are historical relative to this head.

Lab setup

  • Detached worktree of the exact head with its own npm ci and its pinned Electron 44.7.0. The dev daemon was built from this head and the app launched with electron-forge start.
  • Isolated AO_DATA_DIR, run file, Electron profile, HOME, and OPENJARVIS_HOME.
  • uv 0.13.4 installed inside the lab only (sha256 verified). OpenJarvis 1.0.4 installed through AO's in-app installer.
  • No provider key was available. I made no provider calls with fabricated keys, so everything that needs inference is reported as environment-blocked.

Status: exact-head CI is 24/24 passing; the earlier auth.go lint failures are fixed here. GitHub currently reports this head as CONFLICTING with main.

Results

Area Result Observed in the native Electron app or the lab daemon
Harness row without uv PASS Settings → Harness → OpenJarvis (official avatar): "Not installed – uv was not found on PATH. Install uv, then retry." Models: "Install OpenJarvis to see its models."
In-app install PASS "Available via uv tool" → Install → Installing… → Verifying… (~40 s) → v1.0.4, executable ~/.local/bin/jarvis, Authentication Unknown. For a moment after Verifying, the panel showed "jarvis was not found on PATH" before it settled (low).
Setup / onboarding DEFECT After a fresh install with no key or config, Account shows Unknown with no setup action. The UI only offers Set up OpenJarvis (jarvis quickstart) when auth is unauthorized, or as a re-login when authorized/configured. The docs page tells users to run it from Harness settings, but it can't be reached in this state.
Setup output on Apple silicon DEFECT / integration Running jarvis quickstart natively in the lab home (M1) wrote [engine] default="mlx" and default_model="qwen3.5:2b". AO's install plan doesn't include MLX, and the overlay keeps the user's engine. The native fallback then crosses the local/cloud boundary, so every AO session started with "Default engine 'mlx' is unavailable; using 'nim'" · Engine nim · Model qwen3.5:2b. A task in that state would go to NVIDIA NIM with a local model id. Quickstart itself sent its built-in test prompt ("Say hello in one sentence.") to NIM without any key and got a 404, yet still printed "Setup complete!". That was the only prompt any part of this test sent to a model endpoint, and it was rejected; there was no authenticated inference at all.
Models (no key) Consistent with scope Native jarvis model list: 83 nim rows, 0 cloud rows. AO: "openjarvis model discovery returned no models"; the New Task model menu only offers "Refresh models".
Task launch (TUI) PASS New Task (agent picker "OpenJarvis · Auth unknown"), empty prompt → terminal session even though the app's default interface is Chat. argv: jarvis chat --agent orchestrator, under ao agent-process supervise, cwd = session worktree. The tab showed Idle at the native You>.
Programmatic input refusals PASS ao send with multiline input, /help, a You> marker, a Confirm: marker, leading whitespace, 3001 bytes, or a tab each returned AGENT_INPUT_REJECTED with a specific reason. The native composer stayed empty.
Non-default permission mode DEFECT (safe direction) With project permissions=bypass-permissions, the spawn was accepted. argv was unchanged and the overlay [security] was identical to native (enforce_tool_confirmation=true). The docs say only the default mode is supported, but ErrAgentPermissionModeUnsupported is never returned. This is a documentation/behaviour mismatch in the safe direction: no safety bypass was observed, and native confirmations stay enforced.
Chat mode Unsupported, correctly refused ao spawn --mode chat → SESSION_MODE_UNSUPPORTED: openjarvis has no chat driver; no session was created. The terminal tab menu offers no "Switch to chat UI".
Kill / archive PASS Archive → confirm → POST /kill 200. The session's pty-host, supervisor and jarvis exited and its clean worktree was removed. A second OpenJarvis session was untouched (same PIDs, still idle).
Restore Unsupported, correctly refused POST /restore → 409 AGENT_RESTORE_UNSUPPORTED "openjarvis native chat history is in memory; start a new session explicitly", shown inline; no agent or pty process started. Low: the refused restore recreated the clean session worktree and left it behind.
Config overlay Note data/agents/openjarvis/<sid>/config.toml (mode 0600) copies 10 of 11 native config sections verbatim and adds the AO instruction prefix and the code-assistant tools. Whatever the user keeps in config.toml gets duplicated into AO data. The lab config held only empty placeholders, so this is a design note, not a demonstrated secret leak.

Environment-blocked or not run (no provider route):

  • a real task with a follow-up and its file effects
  • whether AGENTS.md project instructions are actually consumed
  • the native Confirm: approve/reject flow, and send-while-active refusal
  • active-turn cancellation (Ctrl-C); not advertised by the PR's docs either
  • models/auth with a real key, and the rejected-key "not connected" state
  • completing a provider switch (needs a second installed, signed-in agent; the switch entry is Developer Mode only)
  • the update/version-refresh path

Lab-only environment notes (not this PR):

  • predev's wake-helper Swift compile fails on this host's Command Line Tools (Swift 5.8.1 / SDK 13.3), so the app was launched with the AGENTS.md lab recipe instead.
  • The lab app opened an unrelated gh auth login panel at startup; I closed it without signing in.

Footage: pr6537-openjarvis-lifecycle.mp4 is renderer-only content from the isolated Electron window, captured via CDP screencast: no OS window frame, no audio, 31.6 s, 80 real frames, all decoded and privacy-reviewed. It shows New Task → OpenJarvis REPL (mlx→nim fallback banner) → Archive → "TERMINAL ENDED / [process exited]" → Restore → the 409 refusal. The screenshots show the uv-missing state, the installed state with Unknown auth and no Setup action, and the session banner.

pr6537-openjarvis-lifecycle.mp4

openjarvis-settings-uv-missing

openjarvis-installed-auth-unknown-no-setup

openjarvis-session-mlx-to-nim-fallback

@coderabbitai

coderabbitai Bot commented Nov 12, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

@github-actions

Copy link
Copy Markdown
Contributor

🏆 Review leaderboard

Nov 5–12, 2026 · UTC

Rank Reviewer PRs reviewed Review rounds PR comments
🥇 @ronishrohan @ronishrohan 91 160 14
🥈 @neversettle17-101 @neversettle17-101 11 15 4
🥉 @illegalcall @illegalcall 11 12 4
4 @Meraj-08 @Meraj-08 10 15 0
5 @nikhilachale @nikhilachale 7 19 3
6 @Prasad-D-Ware @Prasad-D-Ware 7 12 2
7 @Annieeeee11 @Annieeeee11 6 9 0
8 @mohakchakraborty2004 @mohakchakraborty2004 6 7 2
9 @Pritom14 @Pritom14 5 8 1
10 @Rishet11 @Rishet11 5 6 3

Ranked by distinct external PRs reviewed, then review rounds, then PR comments. Self-activity and bot activity are excluded.

Show 6 more reviewers
Rank Reviewer PRs reviewed Review rounds PR comments
11 @somewherelostt @somewherelostt 3 11 1
12 @ApexYash11 @ApexYash11 2 6 2
13 @Vaibhaav-Tiwari @Vaibhaav-Tiwari 2 3 0
14 @codebanditssss @codebanditssss 2 2 8
15 @harshitsinghbhandari @harshitsinghbhandari 2 2 0
16 @MohamedKamran @MohamedKamran 1 2 0

@milesatao-svg

Copy link
Copy Markdown
Contributor Author

Account migration verification

  • New PR author verified as milesatao-svg; submission is a draft.
  • Original PR #6537 and this PR both point to 1a9055a5dd2256a87e6002d98f6e3f80b3eb2fac.
  • Published change-count header verified against this PR's actual GitHub file diff.
  • All 10 original human comment bodies and source links reproduced across 1 historical-evidence comments.
  • All 10 original screenshot/recording URLs reproduced and checked as HTTP 200 media.
  • Current GitHub mergeability: conflicts with main; resolution pending.
  • Fresh CI validation and current-head harness certification are pending. Copied test outcomes retain their original run/commit provenance.

@OrcInc OrcInc added comp/daemon Go daemon, process lifecycle, and backend control plane. enhancement New feature or request labels Nov 12, 2026
@OrcInc OrcInc added this to the Agents & orchestration milestone Nov 12, 2026
nikhilachale and others added 2 commits November 14, 2026 22:52
Resolve conflicts against main adf9f6a:
- drop the cloud/internal/worker/script_steps_test.go edit; the control
  plane left this repository in OrchestratorInc#7352
- keep main's scriptStepTimeoutUnit rename, update-notice test and
  prompt-readiness loop; re-add openjarvis to the CLI, DTO, installer,
  OpenAPI and generated schema harness lists
- renumber the unshipped OpenJarvis migration to 20261114171639069 so it
  follows main's newest migration, and base its test there

Integrate with main's newer harness surfaces:
- request rendered-screen prompt readiness through main's RenderedScreen
  hint; the adapter had dropped the old RequiresRenderedActivity hook, so
  the unterminated "You> " composer was never matched
- reuse main's renderedReadinessRuntime fake instead of redeclaring it
- offer the provider keys the native cloud engine reads (OrchestratorInc#7312)
- give OpenJarvis a neutral Usage tone so AGENT_HUES stays exhaustive

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…diff

- require a whole `jarvis --version` line to equal the pinned
  "jarvis, version 1.0.4" instead of a substring match, which also
  accepted 1.0.40 or a lookalike `jarvis`; give `ao doctor` the same
  expected version prefix
- correct the docs: every AO permission mode runs with native security,
  and provider keys can be saved for AO in Harness settings
- rename the avatar licence to the LICENSE-<id>.txt convention
- drop the committed research reports, receipts and screenshots; they
  remain at the original-commit links in the PR description

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@nikhilachale

Copy link
Copy Markdown
Collaborator

Current-head update and ao-agent-harness receipt (2026-11-14 17:34 IST)

Head: 9a4b21845388a4b55323c83f6c3aad4ce8f126c3 on main adf9f6a303. GitHub: MERGEABLE, merge state BLOCKED (draft, review and CI). No history rewritten; main was merged in.
Skill used: .agents/skills/ao-agent-harness read at exact commit 04c8c95b23d717ccd2bc314e4577a0596b92e840 (PR #7081, open, not on main). No skill files, contracts, reports or screenshots are in this diff.

What changed in this update

  • 53f254c8f1 merge of main. Conflicts resolved in spawn help, the DTO, installer and schema enums, session-manager tests, the migration ledger, docs nav and the update-notice test. The obsolete cloud/internal/worker/script_steps_test.go edit is dropped: the control plane left this repo in chore: move the AO Cloud control plane to its own private repository #7352, so main's behaviour is unchanged.
  • Migration renumbered to 20261114171639069_allow_openjarvis_harness.sql, above main's newest (20261114072016988). No main migration after the old ID rebuilds sessions or touches its harness CHECK. The SQL is idempotent (NOT LIKE '%openjarvis%'), and its up/down test now starts from main's newest version.
  • Prompt readiness was dead at the old head: the adapter had dropped RequiresRenderedActivity, so the unterminated You> row was never read and startup waited the full 60s. It now uses main's RenderedScreen hint.
  • The OpenJarvis readiness test redeclared main's renderedReadinessRuntime and would not compile after the merge. It now reuses main's fake and the adapter's real hints.
  • Wiring for main surfaces added after the PR's base. Without these, CI fails after the merge:
  • 9a4b218453 skill fixes:
    • exact jarvis, version 1.0.4 line match instead of strings.Contains, which also accepted 1.0.40 and lookalikes, plus a unit test
    • ao doctor version prefix for the generic jarvis binary
    • licence renamed to LICENSE-openjarvis.txt
    • docs corrected on permission modes and stored keys
    • committed research reports, receipts and screenshots removed from the diff; they stay at the original-commit links in the description

Receipt (per skill step)

Step Result Evidence or gap
0 Intake PASS (manual) This open PR is continued. harness_intake.mjs was not run: no execution on the worker's Windows host
1 Compatibility PASS (source review) Binary jarvis, pinned 1.0.4. Install: uv tool install OpenJarvis[...]==1.0.4. Setup: jarvis quickstart. Auth: real provider round-trip (configured when inconclusive). TUI: jarvis chat --agent orchestrator, prompt after start. System prompt: per-session config overlay (guide channel 3). Models: jarvis model list catalog plus direct entry. Unsupported, documented: Chat/ACP, native session ID, restore (history is in memory only; typed 409 AGENT_RESTORE_UNSUPPORTED), hooks, per-mode permissions (native security in every mode), multiline input. Not re-verified live at this head
2 Touchpoints FAIL, 1 required TODO (justified) Every required TUI, install, auth, models and branding row is wired, checked by hand against the script's predicates (script not run). activity-dispatch has no entry: the CLI has no hooks and status comes from DetectTerminalActivity, the same as Hermes. Needs reviewer acceptance; the check was not weakened. Optional and absent: docs/harnesses/openjarvis.md. Chat and reviewer are not integrated, by design
3 Contract BLOCKED local-contracts.json belongs to the skill PR. Proposed entry: binary: jarvis, versionArgs: ["--version"], surfaces: ["tui"], adapters.tui: openjarvis. No integration (there is no login-status command), no restoredReady (no restore). models: ["model","list"] prints a Rich table that neither the json nor the lines format parses (a skill-side gap). doctor not run
4 Live audit BLOCKED Not run: Windows worker, and the VPS is off-limits. Expected by design: tui_native_id FAIL and tui_restore FAIL (409), tui_restore_history SKIP, Chat group SKIP. tui_activity relies on terminal detection (up to about 30s lag) and needs live proof. Multiline audit prompts would be refused
5 UI audit BLOCKED No desktop/CDP run on this host. All 9 checkpoints are open
6 Hand-off BLOCKED Waiting on steps 4–5

CI

All four workflows on this head (CI, Public boundary, gitleaks, Harness release feeds) are action_required: fork-PR runs need a maintainer to approve them. No local build, lint or tests ran (Windows host policy).

Open risks

  • conpty/surface.go resize change is shared code (no build tags, so every OS and every TUI harness). Only generic tests exist. The guide asks for Codex and Claude Code coverage, and its effect on inline TUIs after a shrink needs a decision and live proof.
  • The You> ready cue is not bottom-anchored. Every write is still re-validated against the anchored, current composer by daemon admission, so the worst case is a refused send.
  • charmbracelet/ultraviolet is imported directly but still marked // indirect in go.mod (hygiene only; CI has no tidy check).

Not merge-ready: live TUI audit, UI audit and approved CI are all still required.

@nikhilachale nikhilachale mentioned this pull request Nov 14, 2026
4 tasks done

@nikhilachale nikhilachale left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Changes requested. Reviewed PR #8205 at 9a4b218453 against main. The adapter, migration, reconcile repair, install plan, catalog parser and API/UI registration follow the existing Hermes-style harness conventions. Four required changes are inline:

  1. The submission witness only accepts a screen that ends at the echoed line. Any native output after Enter makes a delivered send report unconfirmed, and AO input stays locked until kill/restore.
  2. The agent-failure marker check in DetectTerminalActivity marks the session Blocked whenever the marker is anywhere in the 40-line viewport, including after a fresh empty You> prompt.
  3. AuthStatus reads provider keys with os.Getenv, bypassing the readiness ctx overrides (authprobe.Getenv), so the reported status can disagree with what sessions receive.
  4. ErrAgentPermissionModeUnsupported is never returned, so its mapping and TestEveryPermissionModeUsesNativeDefaults are vacuous.

Optional, not required:

  • Consider reusing the authprobe redirect/scheme guards in probeProviderKey. In net/http, custom x-api-key / x-goog-api-key headers survive cross-host redirects.
  • openjarvis.png is about 393 KB; a downscaled asset would be lighter.
  • The LOGOS entry is placed first instead of in the surrounding order.

The shared conpty/surface.go resize change still needs the maintainer decision and cross-harness live proof the PR already lists. This was a read-only review: I did not run tests or builds.

Comment thread backend/internal/ports/agent.go Resolved
… auth

- confirm a submission where the baseline's empty composer was, aligned
  below the baseline history, and ignore the turn output and new prompt
  printed after the echo; before, any output after Enter left the send
  unconfirmed and blocked AO input until kill/restore
- judge the agent-failure marker only after the last prompt, so a fresh
  empty prompt below an earlier failure is idle again
- read provider keys and OPENAI_BASE_URL through authprobe.Getenv, so a
  key AO withholds from OpenJarvis sessions is not probed
- remove the never-returned ErrAgentPermissionModeUnsupported and its
  mapping; the permission test now stubs the binary and asserts the same
  argv for every mode

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@nikhilachale nikhilachale left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approved. I re-reviewed PR #8205 at 5e109bdf47, looking at the fix commit since 9a4b218453. All four requested changes are resolved:

  1. Submission witness: TerminalMessageObserved now lines the screen up with the baseline history above the composer that was empty, then requires You> <message> at that row. Rows printed after the echo and the next empty prompt are ignored, so a delivered send no longer locks AO input. Repeated history picks the least-scrolled match and fails closed. New tests cover output after the echo, a new prompt, scrolled history, a padded echo row, and an identical earlier turn.
  2. Failure marker: the marker now only counts after the last column-zero prompt, or anywhere when there is no prompt. A fresh empty prompt reads as Idle again, and the submitted line itself is excluded.
  3. Auth environment: every key and OPENAI_BASE_URL lookup on the auth path goes through authprobe.Getenv(ctx, ...). A new test covers routed and unrouted keys that AO withholds.
  4. Unused sentinel: ErrAgentPermissionModeUnsupported, its mapping and its test row are removed, with no references left. The permission test stubs the binary and checks the same argv for every mode.

Optional, not required:

  • A long submitted line that wraps can put the failure marker on a continuation row below the prompt, which would read as Blocked. Reserving that marker in validateLine, as is already done for You> and Confirm:, would close this.
  • The earlier optional notes still apply: the authprobe redirect guards, the asset size, and the LOGOS order.

The shared conpty/surface.go resize change and the live TUI/UI audits are still open PR checklist items. They are outside this code review. This was a read-only review: I did not run tests or builds.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

comp/daemon Go daemon, process lifecycle, and backend control plane. enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants