You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
plugins: decide tool-restriction behavior in Cursor, which ignores disallowed-tools and agent tool lists #12483
Cursor does not enforce our tool restrictions. In Cursor a skill or plugin agent that Claude Code limits can still use the tools it was meant to be denied.
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Small
Complexity: Low
Uncertainty: Known
Flags: none
Blocked by: none
Decision recorded.
Agent choice, not an owner answer.
This records the choice Lane 2 made in comment 6588070692 on 2026-11-02T20:47Z, so the issue can return to admission.
[quote truncated; the full decision is at the link]
That decision settles the open questions. Tool limits are documented as Claude Code-only in docs/cursor-compatibility.md, no Cursor readonly field is added, and the Q561 plan states that only Claude Code enforces it. This is a bounded, docs-only change. I found no duplicate among the open issues.
Refinement. A read-only run explored this repository, searched its issues and pull requests, and read the allowed docs to draft what follows. On an issue a listed human filed, it counts as the issue's spec once posted, while unedited; it is still model text, so check it.
Intended outcome: Tool limits (disallowed-tools, disallowedTools, agent tools:) are documented as Claude Code-only, with a hook-backed read-only check for the reviewer agents only if one is shown to hold in Cursor.
Which approach should Cursor users be told to rely on: treat tool limits as Claude Code-only and document that, back read-only cases with a PreToolUse hook, or both?
RECOMMENDED: Document tool limits as Claude Code-only. Add hook backing for the read-only reviewer agents only after a probe shows a PreToolUse deny holds in Cursor. Basis: judgment
Is Cursor's readonly agent field allowed, even though it is a Cursor-specific key?
RECOMMENDED: No. Keep one plugin source and leave the field out. Basis: judgment
Where should the decision be recorded for the day-job rollout to read it?
RECOMMENDED: Add a Cursor section to docs/upstream/claude-code.md, or to the rollout doc if one exists. The issue does not name the file, so the owner should pick it. Basis: .base/docs/upstream/claude-code.md:39
If the hook route is chosen, which agents count as read-only and must be hook-backed: all 31 agents with tools: allowlists, or only the reviewers in review, planning, testing, multi-agent and discovery?
RECOMMENDED: Only the read-only reviewers, plus the two disallowedTools agents (researcher, intent-tracer). A hook for every allowlist is too broad. Basis: judgment
Does per-job disallowed-tools use (Q561) proceed on the assumption that it holds only in Claude Code?
RECOMMENDED: Yes. Q561 should state that it is Claude Code-enforced only. Basis: judgment
Sources read
/home/runner/work/claude-code-plugins/claude-code-plugins/refine-data/context.json: the issue body gives the Cursor probe result, the affected skills and agents, and the three options. No **Decision recorded.** comment exists.
.base/docs/upstream/claude-code.md:39: the upstream-behavior table is the likely place for a recorded Cursor limitation.
.base/plugins/harness-config/CHANGELOG.md:305: audit-pass uses disallowed-tools as a narrowing aid, and the read-only contract is instruction-held.
To answer, post a comment that starts with this line:
**Decision recorded.** <answer>
Automated refinement by the lanes App. Answer with a comment whose first line is **Decision recorded.**.
Agent choice, not an owner answer. The owner did not answer this question. Lane 2 chose on the #7556 evidence, and Issue Triage owns the question. Accept the triage recommendations, with one placement choice. (1) Document tool limits (disallowed-tools, disallowedTools, agent tools:, skill allowed-tools) as enforced by Claude Code only. Add PreToolUse hook backing for read-only agents only after a probe shows a PreToolUse deny holds in Cursor. (2) No Cursor-specific readonly field: one plugin source. (3) Record it in one marketplace-level Cursor compatibility note, the same file #12484 creates (docs/cursor-compatibility.md), so Cursor facts have a single home. (4) If hook backing is later added, it covers the read-only reviewer agents plus researcher and intent-tracer, not all 31 allowlists. (5) Q561's per-job disallowed-tools states that only Claude Code enforces it. Basis: #7556 probe (paired runs showed both keys ignored). Decided under the owner's standing direction (decide from evidence; no intent, security or cost question here: these limits scope agent behavior and are not a secret or token boundary).
This records a re-fire of an earlier refusal, not a new refusal. The lanes kill switch stopped run 41522469600, the work-resolve-issue run that agent-ready started, before it did any work; that run ended at 2026-11-02T20:52:24Z. work-unblock has now removed and added agent-ready back to start a fresh run. A kill-switch refusal is no fault in this issue or the lanes, so it never counts against later re-fires; after limits.limit-stops-per-issue of them since a person last added agent-ready or a run last reported, work-unblock waits longer before each one, up to 24 hours.
Activity
github-actions commentedon Nov 2, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Small
Complexity: Low
Uncertainty: Known
Flags: none
Blocked by: none
#12483 (comment)
That decision settles the open questions. Tool limits are documented as Claude Code-only in
docs/cursor-compatibility.md, no Cursorreadonlyfield is added, and the Q561 plan states that only Claude Code enforces it. This is a bounded, docs-only change. I found no duplicate among the open issues.Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Nov 2, 2026
The latest triage verdict holds nothing: the issue is at
status: ready.melodic-automation-lanes commentedon Nov 2, 2026
Refinement. A read-only run explored this repository, searched its issues and pull requests, and read the allowed docs to draft what follows. On an issue a listed human filed, it counts as the issue's spec once posted, while unedited; it is still model text, so check it.
Intended outcome: Tool limits (
disallowed-tools,disallowedTools, agenttools:) are documented as Claude Code-only, with a hook-backed read-only check for the reviewer agents only if one is shown to hold in Cursor.Related work
Open questions
RECOMMENDED: Document tool limits as Claude Code-only. Add hook backing for the read-only reviewer agents only after a probe shows a PreToolUse deny holds in Cursor. Basis: judgment
readonlyagent field allowed, even though it is a Cursor-specific key?RECOMMENDED: No. Keep one plugin source and leave the field out. Basis: judgment
RECOMMENDED: Add a Cursor section to docs/upstream/claude-code.md, or to the rollout doc if one exists. The issue does not name the file, so the owner should pick it. Basis: .base/docs/upstream/claude-code.md:39
tools:allowlists, or only the reviewers in review, planning, testing, multi-agent and discovery?RECOMMENDED: Only the read-only reviewers, plus the two
disallowedToolsagents (researcher, intent-tracer). A hook for every allowlist is too broad. Basis: judgmentdisallowed-toolsuse (Q561) proceed on the assumption that it holds only in Claude Code?RECOMMENDED: Yes. Q561 should state that it is Claude Code-enforced only. Basis: judgment
Sources read
**Decision recorded.**comment exists.disallowed-toolsas a narrowing aid, and the read-only contract is instruction-held.To answer, post a comment that starts with this line:
Automated refinement by the lanes App. Answer with a comment whose first line is
**Decision recorded.**.kyle-sexton commentedon Nov 2, 2026
Agent choice, not an owner answer. The owner did not answer this question. Lane 2 chose on the #7556 evidence, and Issue Triage owns the question. Accept the triage recommendations, with one placement choice. (1) Document tool limits (
disallowed-tools,disallowedTools, agenttools:, skillallowed-tools) as enforced by Claude Code only. Add PreToolUse hook backing for read-only agents only after a probe shows a PreToolUse deny holds in Cursor. (2) No Cursor-specificreadonlyfield: one plugin source. (3) Record it in one marketplace-level Cursor compatibility note, the same file #12484 creates (docs/cursor-compatibility.md), so Cursor facts have a single home. (4) If hook backing is later added, it covers the read-only reviewer agents plusresearcherandintent-tracer, not all 31 allowlists. (5) Q561's per-jobdisallowed-toolsstates that only Claude Code enforces it. Basis: #7556 probe (paired runs showed both keys ignored). Decided under the owner's standing direction (decide from evidence; no intent, security or cost question here: these limits scope agent behavior and are not a secret or token boundary).melodic-automation-lanes commentedon Nov 2, 2026
This records a re-fire of an earlier refusal, not a new refusal. The lanes kill switch stopped run 41522469600, the work-resolve-issue run that
agent-readystarted, before it did any work; that run ended at 2026-11-02T20:52:24Z. work-unblock has now removed and addedagent-readyback to start a fresh run. A kill-switch refusal is no fault in this issue or the lanes, so it never counts against later re-fires; after limits.limit-stops-per-issue of them since a person last addedagent-readyor a run last reported, work-unblock waits longer before each one, up to 24 hours.13 remaining items
github-actions commentedon Nov 3, 2026
github-actions commentedon Nov 3, 2026