Skip to content

feat: make data sources, test scope and commands work for any repo - #16922

Merged
kyle-sexton merged 3 commits into
mainfrom
feat/skills-serve-any-repo
Nov 17, 2026
Merged

kyle-sexton merged 3 commits into
mainfrom
feat/skills-serve-any-repo

Conversation

@kyle-sexton

Copy link
Copy Markdown
Member

No related issue: workshop 06 close-out, owner decisions Q762-Q766 (2026-11-17)

Summary

Five owner-approved changes so the app-control, toolchain and mcp-tools skills work in any consumer repository, whatever its stack. Stack specifics come from the consumer repository at run time, never from a per-stack example or default.

Fix

  • Q762, app-control: a sources entry read with a credential names the credential's variable (credential) and a consumer-supplied read-only grants command. /app-control:map asks for both with the read level; /app-control:drive ready runs the command where the entry's level allows and reports the exit. A source that offers no scope control records grants: full-user-access: it has no read-only grant to show. That source, and one with a credential and no passing grants record, is capped at attended, and ready names the reason. The contract states the basis (a read-only rule held only in the agent's tool can be bypassed, so the guarantee rests on the store's own grants) and points at ADR 0086.
  • Q763, app-control: an optional data field records the data category a source exposes, who may read it, and the review that decided both. The control skill records the review; it does not perform it. Map asks for it before any read level above never; ready lists it.
  • Q764, ecosystem-commands 1.6.0 and toolchain: an optional test-changed-cmd key, a change-scoped test command over <files> (also a requires key). /toolchain:check runs it for a change and the full test-cmd under all or with no changed files; a repository's CI calls the same recorded command. /toolchain:setup drafts it from the repo's own test runner, or through /discovery:research when the repo has none. No bundled default changes.
  • Q765, mcp-tools: info-level criterion C21. A tool whose destructiveHint is true has a preview or dry-run counterpart (a tool or a mode) that runs the same checks and changes nothing. It points at the --dry-run trait in the app-control control-skill contract. Criteria counts move from 20 to 21 and from eight categories to nine.
  • Q766, toolchain: /toolchain:setup drafts each command key from the repo's own task runner or wrapper scripts before the bundled default, in any ecosystem. This general rule replaces the one-ecosystem scripts special case, which becomes one instance of it. No new key for other verbs.

Verification

  • scripts/validate-plugins.sh --only for app-control, toolchain and mcp-tools: all manifests and the catalog validated.
  • docs/contracts/ecosystem-commands/examples.test.sh: PASS=10 FAIL=0. scripts/check-contract-clause-coverage.py: passed. scripts/check-standards-contract-bump.sh origin/main: no bumps required.
  • scripts/check-changelog-fragments.sh --check-required origin/main: every fragment-mode plugin changed (3) has a fragment. No eval or test pins the changed text; no eval was touched.

Related

  • Owner decisions Q762-Q766, 2026-11-17.
  • ADR 0086 (read tier bounded by a read-only principal).

🤖 Generated with Claude Code

kyle-sexton and others added 2 commits November 17, 2026 14:01
- app-control: a data source read with a credential names its variable
  and a consumer-supplied read-only grants command; drive ready runs it
  and caps a source without a passing record at attended.
- app-control: a data source may record its data category, readers and
  the review that decided them; map asks before a level above never.
- ecosystem-commands 1.6.0: optional test-changed-cmd, a change-scoped
  test command over <files>; toolchain:check runs it for a change and
  test-cmd under all; setup drafts it from the repo's own runner.
- toolchain:setup drafts command keys from the repo's own task runner
  or wrapper scripts before the bundled default, in any ecosystem.
- mcp-tools audit: info-level C21, a destructive tool has a preview or
  dry-run counterpart that runs the same checks and changes nothing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A source that offers no scope control records grants: full-user-access.
It stays capped at attended, like a source with no grants record, and
drive ready names the reason.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@kyle-sexton
kyle-sexton marked this pull request as ready for review November 17, 2026 19:05
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Nov 17, 2026 •

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-17T19:07:09.002187Z dcebb4b Draft marked ready
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.

@github-actions

Copy link
Copy Markdown
Contributor

Lane run report

Field Value
Activity pr-refine/fix-docs
Model rung 0
Leg handoff
Run job success
Activity step success
Commits pushed none (read lane)
Run 44154873180

Failed GitHub API calls

Method Path Status
GET /repos/{repo}/git/ref/tags/* 404

@github-actions

github-actions Bot commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

claude-security-review reviewed 4accb9a8a22c92d789045ddf8f52bdd3043f41fb: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Security review (incremental, blocking only): no findings

This review covers only the change since dcebb4be, which touches one file: plugins/mcp-tools/skills/audit/reference/checklist.md. I read its whole hunk. This lane already reviewed an earlier head, so the review was blocking only.

What changed: C21 is an info-level audit criterion that asks whether a destructive tool has a preview counterpart. It now applies only when a tool's readOnlyHint is not true and its destructiveHint is true. A tool marked read-only now passes. This matches how MCP ToolAnnotations read destructiveHint.

Why it is not a security finding:

  • The change is advisory checklist text. It narrows which tools get an info-level note and removes no guardrail.
  • It touches no trusted path or trust control: the trusted-actor list, token broker, kill switch, merge holds and .github/actions/ are all unchanged.
  • No plugin.json or marketplace.json changed, so there are no manifest lines.

The fix is the one the settled thread on this file asked for, and it was answered by the owner.

critical-findings: 0

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

claude-review: 1 inline finding(s) on dcebb4becc6ec5915aae2dd27d4b67d715dd9013; the summary is in this lane's pull request comment.

claude-lane: claude-review

@github-actions

github-actions Bot commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

claude-review reviewed 4accb9a8a22c92d789045ddf8f52bdd3043f41fb: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Code review: PR #16922 (incremental, blocking only)

Scope: what changed between dcebb4b and 4accb9a. That is one hunk in plugins/mcp-tools/skills/audit/reference/checklist.md, the C21 row, and I read all of it. This lane had already reviewed an earlier head, so this review reports blocking findings only.

Result: no new findings. The commit fixes the one finding from the last review (C21 flagged read-only tools).

  • C21 now applies only to a tool whose readOnlyHint is not true and whose destructiveHint is true, either declared or by C13's default.
  • The evaluation cell now says a read-only tool passes, and that destructiveHint counts only when readOnlyHint is false. That is the MCP ToolAnnotations rule.
  • "Not true" in the trigger and "false" in the evaluation cell mean the same thing, because the spec treats a missing readOnlyHint as false.
  • So a list_items tool that declares readOnlyHint: true and omits destructiveHint now passes C21. That also matches C12 and C13 at checklist.md L126–L127.

Settled thread: the owner resolved that thread with a reply stating the fix. I have not raised it again.

Not checked:

  • No closing issue, so there were no acceptance criteria to check.
  • The owner says validate-plugins --only mcp-tools passes. That is their claim; this review ran nothing.

Loosened-check verdict: fail (default; the job supplied no settings); patterns: 0. The changed row narrows which tools C21 flags. It is not a loosening, because the tools it stops flagging are ones the spec says are not destructive.

Verifier dropped 0 of 0 candidates.

…rion

C21 resolved a missing destructiveHint to true through C13's default,
so every read-only tool was reported as lacking a preview counterpart.
The MCP ToolAnnotations schema counts destructiveHint only when
readOnlyHint is false; C21 now requires readOnlyHint to be not true.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@kyle-sexton
kyle-sexton added this pull request to the merge queue Nov 17, 2026
Merged via the queue into main with commit 73fc0c9 Nov 17, 2026
49 checks passed
@kyle-sexton
kyle-sexton deleted the feat/skills-serve-any-repo branch November 17, 2026 19:43
@github-actions github-actions Bot added the merge-path: owner-agent Post-merge accounting record (README.md, Merge paths) label Nov 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

merge-path: owner-agent Post-merge accounting record (README.md, Merge paths)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant