Skip to content

feat(providers): add supervised MCode and DSH ACP previews - #3

Merged
Hylouis233 merged 5 commits into
mainfrom
feat/mcode-dsh-acp
Sep 16, 2026
Merged

Hylouis233 merged 5 commits into
mainfrom
feat/mcode-dsh-acp

Conversation

@Hylouis233

@Hylouis233 Hylouis233 commented Sep 15, 2026 •

Copy link
Copy Markdown
Owner

Summary

Implement the first source-only OpenT3Code Alpha with independent MiniMax Code and DeepSeek Harness ACP drivers, an installable DSH opent3code profile bundle, and a separate supervised source-preview data directory. The user explicitly requested both integrations and a GitHub release.

The adapters support text prompts, assistant/reasoning/tool events, one-shot permission requests, cancellation, scoped process cleanup and instance/workspace/home-bound native resume. MCode's permissionMode option is explicitly set to default (Ask), independently of its Default/Plan work mode; missing, unsettled or incompatible policy fails closed. DSH's companion profile uses workspace-write plus ask and disables session telemetry. Existing native accounts and existing preview settings are not rewritten.

Unsupported attachments, plan-mode changes, native transcript import, rollback, mid-turn steering, background text generation and non-supervised modes are rejected explicitly. DSH's native resume does not replay history. This is not a signed desktop installer, generic terminal scraper, universal CLI bridge or multi-user sandbox.

Release and upstream isolation

A version manifest enables only opent3code-v0.1.0-alpha.1. The publisher waits for successful exact-main CI, builds source ZIP plus DSH TGZ with read-only permissions, checks asset/receipt hashes, uploads a draft, and publishes it as a prerelease only after rechecking CI and asset completeness. No npm publication, desktop signing or upstream v* tags are used. Inherited production/preview publishers are restricted to the upstream repository. Main protection remains unchanged.

Evidence

  • Strict read-only verification: https://github.com/Hylouis233/opent3code/actions/runs/34992090870 — 24 ACP regression tests, 50 maintenance/launcher/release-policy tests, affected-workspace typechecks, lint and unused-code checks passed.
  • Actual MCode 0.4.7: initialization, unauthenticated-session rejection, session creation with an isolated loopback provider fixture, and an actual transition from inherited bypassPermissions to permissionMode=default were observed. No paid account or production model was used. The local generation experiment did not complete due to fixture model-routing configuration; it is NOT reported as a successful generated turn.
  • Actual DSH CLI 0.1.5-rc.1 / profile dependencies 0.1.5-rc.2: companion package installation, effective ask/workspace-write/telemetry-disabled configuration, initialization and new native session creation succeeded. Installation initially exposed pnpm's native-build approval requirement; the companion now initializes a dedicated policy with exact observed native versions instead of allowing every build script.
  • Controlled ACP subprocess tests cover complete text turns, approvals/refusals, cancellation, crash handling, scoped resume and unsafe/missing permission metadata. These are not a live paid-account or all-platform GUI certification.

Full PR CI must pass on the current candidate before merge, then exact-main CI before release. Earlier failed diagnostic runs or another commit's green checks do not qualify. Source-transfer operations did not bypass branch protection; temporary operations files are not part of this PR.

Implementation: GPT-6 Astra Pro via ChatGPT, local source analysis and GitHub Actions.

Add explicit native driver identities, scoped sessions, one-shot permission mapping, cancellation and resume validation. Include subprocess fixtures. Targeted preflight exposed API/type errors; this commit applies their focused fixes. Full PR CI is required before merge. No release or runtime authentication claim.
@github-actions github-actions Bot added size:XL 500-999 effective changed lines. vouch:trusted Author has write access or is explicitly trusted by OpenT3Code. labels Sep 15, 2026
…view launcher

Package the official ACP profile as an independently installable DSH bundle. Initialize new OpenT3Code preview state with supervised permissions without overwriting native tool accounts or existing preview settings. Document account ownership and unsupported capabilities. Native installer identity migration remains out of scope for this source Alpha.
@github-actions github-actions Bot added size:XXL 1,000+ effective changed lines. and removed size:XL 500-999 effective changed lines. labels Sep 15, 2026
@github-actions

github-actions Bot commented Sep 15, 2026 •

Copy link
Copy Markdown

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider Metric Main baseline This PR Impact PR ceiling
Codex Total thread wire 13.6 KiB 13.6 KiB −7 B (−0.1%) 15.1 KiB ✅
Codex Thread snapshot wire 7.1 KiB 7.0 KiB −9 B (−0.1%) 7.3 KiB ✅
Codex Live turn WebSocket wire 6.6 KiB 6.6 KiB +2 B (+0.0%) 7.8 KiB ✅
Codex Live turn WebSocket decoded 57.1 KiB 57.1 KiB 0 B (0.0%) 66.4 KiB ✅
Codex Live turn messages 10 10 0 (0.0%) 21 ✅
Claude Total thread wire 13.6 KiB 13.6 KiB +16 B (+0.1%) 15.1 KiB ✅
Claude Thread snapshot wire 7.1 KiB 7.1 KiB +2 B (+0.0%) 7.3 KiB ✅
Claude Live turn WebSocket wire 6.5 KiB 6.5 KiB +14 B (+0.2%) 7.8 KiB ✅
Claude Live turn WebSocket decoded 57.8 KiB 57.8 KiB 0 B (0.0%) 66.4 KiB ✅
Claude Live turn messages 9 9 0 (0.0%) 21 ✅

Baseline: 712bace · PR result: d0f3fa5 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

  • Codex decoded thread snapshot: 113.9 KiB
  • Claude decoded thread snapshot: 114.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

…rovals

Pre-create only the dedicated profile's pnpm policy, preserve existing profiles, and allow the exact native dependency versions observed during install. Keep all-build execution disabled. Move Node launcher tests outside the Vitest workspace to avoid mixing test runners.
Publish only an explicitly enabled source prerelease after the precise main push passes every required job. Build without write credentials, verify checksums and receipt, upload a draft, and make it public only after completeness and CI are rechecked. Do not publish npm packages, native installers or upstream tags. Include fail-closed policy regression tests.
…rce release

Distinguish MCode permissionMode from default/plan work mode and fail closed on missing or unsettled approval policy. Match actual native CLI metadata with regression fixtures. Normalize source, include launcher/release tests in CI, isolate inherited publishers, and correct source-preview instructions.

Read-only verification run 34992090870: 24 ACP tests, 50 maintenance tests, lint, typecheck and unused-code checks passed. Actual MCode confirmed a transition from bypassPermissions to Ask using a loopback-only fixture. DSH companion installation, effective configuration and native session creation were verified separately. Paid-account generation and native desktop installers are not certified. Full PR and exact-main CI remain required before publication.

Implementation: GPT-6 Astra Pro via ChatGPT and GitHub Actions.
@Hylouis233
Hylouis233 marked this pull request as ready for review September 15, 2026 16:06
Copilot AI lite review requested due to automatic review settings September 15, 2026 16:06
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

@Hylouis233
Hylouis233 merged commit 745f7c4 into main Sep 16, 2026
18 of 20 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

size:XXL 1,000+ effective changed lines. vouch:trusted Author has write access or is explicitly trusted by OpenT3Code.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants