Skip to content

Preserve graph cache isolation, renderer lifecycle and scoped queries - #241

Closed
Coding-Dev-Tools wants to merge 100 commits into
mainfrom
codex/dashboard-cache-readiness-20260928
Closed

Coding-Dev-Tools wants to merge 100 commits into
mainfrom
codex/dashboard-cache-readiness-20260928

Conversation

@Coding-Dev-Tools

@Coding-Dev-Tools Coding-Dev-Tools commented Sep 29, 2026 •

Copy link
Copy Markdown
Owner

Graph callers receive isolated cached scenes for all layers and explicitly empty layers, and concurrent eviction or graph-job completion cannot turn a cache hit into an HTTP 500. Cache metadata changes share a lock; copying and database work remain outside it. Renderer pause state survives WebGL context loss and restoration.

PNG exports reproduce the authored CSS gradients, grid spacing, screen blending, reduced-motion opacity and foreground composition. Fractional display scaling preserves the canvas drawing scale and avoids stretching rounded gradient tiles. An independent browser screenshot supplies the pixel oracle.

Graph classification starts from scoped candidate entities and probes indexed source/destination edges, avoiding scans of unrelated repositories. Workspace-wide privacy for incident edges, time-travel behavior, pre-limit private filtering and traversal caps remain intact. A self-loop contributes one endpoint incidence. Secure erasure preserves a global entity referenced by a retained edge from any workspace. Workspace-owned entities still require exact-workspace incidence, so foreign edges cannot prevent their erasure. The indexed endpoint probes remain in that owned-entity branch.

The original local source snapshots are preserved in this branch and in verified recoverable patches. Stacked on #240; merge the #238 → #239 → #240 stack first.

Validation:

  • Cache identity preserves layers=None separately from layers=[]. Both query orders failed before repair; both now retain correct results on actual warm hits. Independent eight-dispatch probes passed with graph row rebuilding forbidden. The successful-refresh subject retirement fix from Include managed Jev for Pro and Team with current trials and workspace routing #240 is integrated, including persisted and in-process replay prevention; final combined graph/session/evidence/MCP qualification passed all 884 tests.

  • Inherited from Include managed Jev for Pro and Team with current trials and workspace routing #240: All command decisions now keep automatic execution disabled, including overconfident remote advice. Required inputs validate before both local and remote processing; incomplete calls return null conclusions without backend lookup or allowance use. Completion requires state plus goal with action history optional; custom accepts state or question. The README, MCP reference, portable skill and generated contract describe this behavior. Four complete MCP modules passed 592 tests, with 20 final real-client mapping checks; selections overlap. Both review findings are fixed and resolved.

  • The README now documents the included Jev plan benefit, Classic/Smart decision entry points, explicit consent and managed/BYOK configuration. The earlier README-only correction passed all 42 documentation/package/skill checks; the final source is qualified by the fresh immutable evidence below.

  • Chromium lifecycle/export suite for the unchanged renderer: 20 passed, without retries or skips. Includes 11 PNG cases at DPR 1/2 plus Cyber at DPR 1.25. The original 10 PNG checks failed before repair; the fractional case independently reproduced resampling. CSS-to-export channel tolerance remains 3.

  • Final graph/cache selection: 211 passed, including deterministic concurrent clear/eviction and budget regressions. Final v115 evidence, documentation, render, package, skill and MCP selection: 884 passed. Final erasure/store/secret/FTS/tombstone selection: 333 passed, one existing skip. Six global/scoped-edge negative controls failed before repair; the final incidence matrix covers 24 cases. The full asset module passed 261 tests after correcting its CSS-ownership guard.

  • Earlier integrated graph/store/scene/privacy/history and relocation selection: 692 passed, one existing Windows POSIX-permission skip. Selections overlap and are not an additive unique-test count.

  • Deterministic SQLite VM counters reject unrelated-edge scan growth; query plans use endpoint indexes. This is not an end-to-end latency claim.

  • JavaScript syntax, Ruff, commercial manifest, dashboard assets and whitespace checks passed. Final pinned Pyright 1.1.414 project qualification passed.

  • Immutable v115 reruns unchanged fixture aggregates; all 583 source bindings match staged Git blobs. Earlier artifacts remain immutable. Draft v99/v100 source and publication snapshots remain recoverable.

Current head: f10dbc40de2d8bb2b0b7ab9bddb4870f6115a64c. Cache, PNG and compatible global-incidence review findings are addressed and resolved. Strict MCP remote consent and a valid offline discovery example are integrated from #240. All 29 required GitHub checks passed on this revision. No production deployment, provider call or release is included.

claude and others added 8 commits September 30, 2026 02:24
…ures

npm published three advisories after this stack last passed CI, failing both
Pi extension jobs introduced by the full development-tree audit:

- fast-uri <=3.1.7 and ip-address <=10.7.0 are production dependencies of
  the MCP SDK. Update the Pi lockfile to fast-uri 3.1.8 and ip-address 10.7.2.
- brace-expansion <=5.0.11 exists only inside the Pi coding agent used as the
  CI test host. That package ships its own npm-shrinkwrap.json, so npm
  overrides cannot replace it, and no Pi release (through 0.99.1) has a fixed
  lockfile yet.

Production dependencies stay strictly audited. Development advisories still
fail, except those confined to the test host's own lockfile, which are
reported as warnings until a fixed host can be adopted.

The step now runs under bash on both runners. On Windows the default pwsh
wrapper reported only the final command's exit code, so a failing npm ci,
verify or integration run was hidden whenever the audit passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ksPGXPZQTikJYpaGHa6R6
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ksPGXPZQTikJYpaGHa6R6
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ksPGXPZQTikJYpaGHa6R6
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ksPGXPZQTikJYpaGHa6R6
@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.

1 similar comment
@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.

@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.

1 similar comment
@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.

@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.

1 similar comment
@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.

@Coding-Dev-Tools
Coding-Dev-Tools changed the base branch from codex/managed-jev-benefits-20260928 to main October 1, 2026 09:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants