feat(trellis): read-only agent profile view in settings and project pages

[Andrey170170]

Trellis v1 Stage C, Lane T3 step 7: read-only agent profile view (design: Trellis docs/v1/agent-profiles.md §3, §8; API: docs/api.md "Agent profiles").

Change

• RPC trellis.getProfile {target?} (read scope) wraps GET /v1/profile.
  • Decoding is lenient: optional fields, unknown statuses kept as text, unknown fields ignored.
  • An older Trellis gets a message naming the version it needs.
• Settings → Trellis → Agent profile: the profile with no target, i.e. the agent homes and the global layer.
  • A Claude/Codex toggle and a refresh button sit in the section header.
  • Items are grouped into MCP servers, Skills and Instructions; Plugins and Repository settings appear when present.
  • Each row shows its source layer, a one-line detail (transport and URL or command, or the skill's description), and badges: Needs approval, Needs trust, Not enforced, Not loaded, Approved (with who approved it), Error, "Off in ".
  • Profile errors that no row shows are listed, attributed to the provider being viewed.
  • A failed refresh shows a stale notice beside the cached data.
• Project settings (Trellis projects): the same view per workspace (a select when the project has live forks).
  • A t3-code row comes first, since T3 gives its own MCP server to both providers.
  • A trusted or untrusted badge, and when the profile was generated ("sessions opened before then run with the earlier profile"), with a warning when Trellis reports it stale.
• Read-only: no approve, apply or trust controls; those are Stage D.
• docs/user/trellis.md: one sentence.

Left out

"Changed since the session opened" per thread: a session's open time lives only in each thread's own projection, which the project page doesn't load. The generation time and its rule are shown instead. A per-thread badge in the chat view, which has the projection, is a natural follow-up.

Checks

• Tests:
  • Trellis.test.ts decodes a realistic untrusted-workspace view (every special state, an unknown status, profile errors), filtering by provider, and the no-target view.
  • TrellisProfile.logic.test.ts covers grouping, badges (approved + off, with and without a layer) and per-provider error attribution.
• Browser against /trellis/dev-t3 (Trellis 479b13f), with a temporary project-layer profile.toml in an untrusted project, removed afterwards. Screenshots are kept locally: /tmp/stagec-profile-*.png.
• Local CI: knip:check, vp check, vpr typecheck and release-smoke pass. t3 tests fail only on the known CodexInstallation and AcpSessionRuntime.processTree failures. Package tests fail only on the known desktop libsecret and snapshot file-mode failures (desktop rerun after an Electron install race); client-runtime, relay and mobile pass when run separately. build:desktop fails on the known libsecret helper.
• Reviews:
  • Codex gpt-6.1-sol: stale profile after a failed refresh, error attribution across providers, Off badge on approved items; fixed.
  • Codex gpt-6-astra: approved instructions that are off; fixed.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-03T03:57:34.141300Z	0c73f51	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.