Repository navigation
Conversation
added 4 commits
November 18, 2026 16:17
Route egm-lite Goldfish reviews to that skill's stop path instead of a generic child, which is not read-only. Every missing-delegation stop now states what is missing, how to remedy it, and which stages and reports were completed and verified before the stop. Refs google#145
Add --absent to the read-evidence utility. It searches every table and cell of the closed SQLite snapshot in UTF-8, UTF-16, base64, gzip, and zlib forms and never echoes the marker. Compression it cannot decode is reported as inconclusive (exit 3), never as a pass. Refs google#145
Record the manifest placement decision, read-only field evidence per ingestion surface, the pinned loader probe result and negative control, known limitations, a step-by-step live runbook that flags every step with model usage, and the per-gate ledger. All live gates remain unverified, and Cursor stays out of the supported-host table. Remove drifting skill and reviewer counts from the guide, and add an experimental, unverified Cursor note to the harness support status. Refs google#145
Derive expected skills and reviewers from skills/ and agents/ instead of fixed lists. Add manifest path containment, a structural and 10-word overlap zero-duplication check, a static write-probe harness with negative fixtures, a native negative control showing which read-only field the pinned loader honors, and checks that keep the guide free of drifting counts and its gates unverified. Refs google#145
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
sahil1115
marked this pull request as draft
November 18, 2026 11:41
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary of Key Changes
Extends the merged experimental Cursor adapter (design #148, implementation #164) toward the Cursor scope of #145. It adds offline verification and a live runbook. It does not claim support. Installation, workflow, and isolation remain unverified until the live runbook is run. Cursor stays out of the supported-host table. This PR leaves #145 open, and the Grok Bot assessment remains separate.
Key changes:
.cursor-plugin/rules/egm.mdc):egm-lite(added in feat(skills): implement EGM Lite rapid engineering protocol (Stage 2) #438, after feat(cursor): add experimental local Agent support #164) dispatches an ad-hoc Goldfish reviewer with no registered read-only Cursor type. A generic child is not read-only, so the rule routesegm-liteGoldfish reviews to that skill's existing stop path. Every missing-delegation stop now states what is missing, how to remedy it, and which stages and reports were completed and verified before the stop.tests/unit/test_cursor.py):skills/andagents/rather than fixed lists. A cross-check against the Claude adapters'Agenttool keeps the orchestrator exclusion honest..., nothing outside the plugin..cursor-plugin/, plus SHA-256 and 10-word overlap againstskills/,agents/, andresources/.tests/integration/cursor_child_reads.py):--absentscans every cell of the closed child SQLite snapshot for a parent-only canary. It checks UTF-8, UTF-16, base64, gzip, and zlib forms and never echoes the marker. Compression it cannot decode is reported as inconclusive (exit 3), never as a pass.docs/cursor.md): the placement decision and rationale, read-only field evidence per ingestion surface, the recorded probe result, known limitations, a live runbook (installation, isolation canary with per-channel probes, read evidence, write denial, completion records, end to end) that flags every step with model usage, and the evidence ledger. The guide's drifting counts are removed; it said seven skills while eight exist.docs/harnesses.md: one experimental, unverified Cursor note under Support Status. The install table is unchanged.Shared
skills/,agents/, andresources/are unchanged. The Hermes, OpenClaw, and Delta adapters named in #145 are not in the tree: #119 is a requirements design only, so there are no suites for them to regress.Contribution Type
Track: Feature / Architectural Change -- Stage 2: Implementation PR
Uses
Refs #145, notCloses, because the issue's live gates stay open. The design is #148 (merged).Verification
Host and version: Cursor CLI
2026.10.01-e373342, inspected and pinned. No model or Cursor agent session ran.f86e0b4test_cursortest_cursor_child_readstest_copilot_adaptertest_opencode_adapter/test_opencode_trace_verifiertest_windsurftest_claude_traces/test_codex_tracestest_harnesses/test_packaging/test_releaseslinux/arm64(emulated), checkout off86e0b4readonly: truealone registerspermissionMode: default;permissionMode: readonlyalone registersreadonlyruff check tests/On a Windows checkout, 16 existing OpenCode, packaging, and PR-triage tests fail with
WinError 1314(no symlink privilege). This is unrelated to this change and also happens onmain. The Linux results above use a clean clone.Harness evidence:
egm-brainstorm: not run.egm-draft: not run.egm-review: not run.Human Accountability Affirmation