Skip to content

docs: replace restated upstream facts with pointers in knowledge and computer-use #12775

Closed

Description

@kyle-sexton

Refs #12694

Outcome

14 passages in 9 files state a fact an external source owns. Rewrite each as a pointer to the live upstream section, or remove it, so the next upstream change cannot leave a stale copy here. Plugins: knowledge, computer-use.

Done when

  • Every row below is fixed: the restated value is deleted, the local decision stays in this repository's own words, and the pointer goes to the exact upstream section with an as-of date and a recheck trigger, per .claude/rules/skill-bodies-state-current-rules.md. A record beside a restated value does not fix it. A row marked Remove has no local dependency on the fact.
  • The PR passes check-stamps.sh --trigger-less (plugins/attribution/skills/audit/scripts/check-stamps.sh) on the files it touches.
  • Each touched plugin has a changelog fragment.

Scope: the knowledge and computer-use rows of #12694 only: docs/contracts/plugin-reconfiguration/README.md, plugins/knowledge/README.md, plugins/knowledge/skills/distill/SKILL.md, plugins/knowledge/skills/docpage-digest/SKILL.md, plugins/knowledge/skills/docpage-digest/context/anthropic-docs-profile.md, plugins/knowledge/skills/setup/SKILL.md, plugins/computer-use/README.md, plugins/computer-use/skills/diagnose/SKILL.md, plugins/computer-use/skills/setup/SKILL.md. The context-guard and disk-hygiene rows belong to the sibling child.

Rows

knowledge

computer-use

  • plugins/computer-use/README.md:32-34 (r060). Pointer: https://code.claude.com/docs/en/computer-use#screenshots-are-downscaled-automatically Drop the pixel-budget explanation and the claim that smaller monitors do not help; keep the decision that zoom is the lever, and point at the downscaling section. Same text also at: plugins/computer-use/skills/diagnose/SKILL.md:75-77, plugins/computer-use/skills/diagnose/SKILL.md:37-38; a row of its own covers it only if listed.
  • plugins/computer-use/skills/diagnose/SKILL.md:75-77 (r061). Pointer: https://code.claude.com/docs/en/computer-use#screenshots-are-downscaled-automatically Drop the megapixel-count explanation; keep the instruction to read the reference file before advising on resolution, and point at the downscaling section. Same text also at: plugins/computer-use/README.md:32-34, plugins/computer-use/skills/diagnose/SKILL.md:37-38; a row of its own covers it only if listed.
  • plugins/computer-use/skills/setup/SKILL.md:44-46,117-119 (r062). Pointer: none found; Upstream owner of the computer-use MCP tool descriptions (list_granted_applications, request_access); the fetched computer-use page does not name them. Search the docs for a tool reference or confirm via the /mcp tool listing. Drop what list_granted_applications returns and its read-only/no-side-effect description, in both places; keep the decision to use it as the check's probe, with a pointer.
  • plugins/computer-use/skills/setup/SKILL.md:60-61,117-118 (r063). Pointer: https://code.claude.com/docs/en/computer-use#approve-apps-per-session Drop the statement that request_access raises a consent dialog, in both places; keep the rule never to call it during a check, and point at the approval section.
  • plugins/computer-use/skills/setup/SKILL.md:65-66 (r064). Pointer: none found; Upstream doc on whether synthesized input resets the OS idle timer (Windows/macOS input injection or idle-detection docs); the computer-use page does not cover it. If none exists, reclassify as local measurement. Drop the claim that synthesized input does not reset the idle timer; keep the decision to probe and report the three timeouts, and point at an owner source once found. Same text also at: plugins/computer-use/README.md:29-31, plugins/computer-use/skills/diagnose/SKILL.md:42, plugins/computer-use/skills/setup/SKILL.md:79-81; a row of its own covers it only if listed.
  • plugins/computer-use/skills/setup/SKILL.md:104-107 (r065). Pointer: https://code.claude.com/docs/en/computer-use#approve-apps-per-session (confirm the section) Drop the description of the allowlist gate's refusal and its error text; keep reporting focus-stealing utilities as INFO with the retry pattern, and point at the approval section. Same text also at: plugins/computer-use/skills/diagnose/SKILL.md:41, plugins/computer-use/skills/diagnose/SKILL.md:80-81; a row of its own covers it only if listed.

Notes

Drafted by the issue-triage session. Agent choice, not an owner answer. Split from #12694 by whole plugins, so no file is touched by both children; every file above exists on origin/main. Size Medium from the passage count (14 of the parent's 28, which intake sized Large). Three rows (r318, r062, r064) have no pointer found yet and need a docs search before writing; the rule for each is in its row.

Lines are as of 3bd42cabfe (https://github.com/melodic-software/claude-code-plugins/blob/3bd42cabfe6bdef37a05778d396869af0381bb51); locate each passage by its text if main has moved. Each rNNN is that row's number in the order of #7292's "Findings by plugin". Pointers are candidates: fetch each before writing it, and confirm the section still documents the fact.

Related

Activity

  1. added theissue type on Nov 3, 2026
  2. added
    priority: lowNice-to-have, cosmetic, or speculative; opportunistic.
    status: readyTriaged, unblocked, and fully specified; eligible to pick up.
    work-class: scopedA briefed fix or small feature; blast radius bounded by the brief, tests exist.
    on Nov 3, 2026
  3. added
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    on Nov 3, 2026
  4. github-actions commented on Nov 3, 2026

    @github-actions
    Contributor
    No description provided.
  5. melodic-automation-lanes commented on Nov 3, 2026

    @melodic-automation-lanes
    Contributor

    This records a re-fire of an earlier refusal, not a new refusal. The lanes kill switch stopped run 41594365657, the work-resolve-issue run that agent-ready started, before it did any work; that run ended at 2026-11-03T08:38:28Z. 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.

  6. added
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    and removed
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    on Nov 3, 2026
  7. github-actions commented on Nov 3, 2026

    @github-actions
    Contributor
    No description provided.
  8. github-actions commented on Nov 3, 2026

    @github-actions
    Contributor
    No description provided.
  9. melodic-automation-lanes commented on Nov 3, 2026

    @melodic-automation-lanes
    Contributor

    The lanes App's pull request #12822 for this issue was closed without merging. work-unblock removed status: ready and agent-ready, so the hourly triage sweep sends it back to intake triage, which decides what work is left and admits it again.

  10. removed
    status: readyTriaged, unblocked, and fully specified; eligible to pick up.
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    on Nov 3, 2026
  11. 21 remaining items

  12. github-actions commented on Nov 4, 2026

    @github-actions
    Contributor
    No description provided.
  13. github-actions commented on Nov 4, 2026

    @github-actions
    Contributor
    No description provided.
  14. added
    host-runNeeds an unattended run on a host; the host runner takes it, not a person or the issue lane.
    and removed
    agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.
    on Nov 4, 2026
  15. melodic-automation-lanes commented on Nov 4, 2026

    @melodic-automation-lanes
    Contributor

    The work-resolve-issue lane stopped on this issue (host-only): the run reports the work needs a host run the lane cannot make.
    It added host-run and removed agent-ready: an agent run on a host builds it, and it waits on no decision from a person.

    In the run's words:

    host-only: row r318 is the only work left (the recorded decision of 2026-11-04 says #12913 and #12818 delivered every other row). The docs search found no Claude Code docs page that describes the raw .md form or its artifacts (theme={null} fence attributes, hard-tab expansion, &-escaped URLs). Under that decision, the no-page branch removes the sentences in plugins/knowledge/skills/docpage-digest/context/anthropic-docs-profile.md:38-41, and docs/review/skills-and-agents.md requires before-and-after eval evidence for that cut. That evidence needs the claude CLI on a host, so the decision routes this branch to a host session.

    What the run says it did not do:

    • Row r318 removal of the channel-artifact sentences in plugins/knowledge/skills/docpage-digest/context/anthropic-docs-profile.md:38-41, with before-and-after eval evidence per docs/review/skills-and-agents.md. Left to a host session with the claude CLI, per the recorded decision.
    • check-stamps.sh --trigger-less and the knowledge changelog fragment: nothing was edited, so these fall to the host run that makes the cut
    • tidy, review, build and test: skipped because the run made no change

    Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/41765710753

    To hand it back to the lane, remove host-run and add agent-ready again.

  16. self-assigned this
    on Nov 4, 2026
  17. kyle-sexton commented on Nov 4, 2026

    @kyle-sexton
    MemberAuthor
    No description provided.
  18. added
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Nov 4, 2026
  19. kyle-sexton commented on Nov 4, 2026

    @kyle-sexton
    MemberAuthor

    Host run on row r318: the cut does not ship. The negative control fell short, so under rewrite-a-plugin.md the part is unmeasurable and "ships its before text". Nothing is committed or pushed. This is the last row, and "Every row below is fixed" cannot be met as written, so the row needs the owner's call. Lease released.

    Search. The llms.txt index lists no page on the raw markdown form (searched: markdown, raw, theme=, fence). Index only, not page bodies. The no-page branch applies.

    Passes. A new case, dd-raw-md-artifacts-reproduced, plus dd-tagging-rules; sonnet, 3 runs per case, --ablation none, 18 VALID runs.

    Version Mean score New case
    Before 1.00 1.00
    After 1.00 1.00
    Degraded 0.858 0.75

    The mean fell 0.142 against the 0.15 bar, and at 2 cases the noise report can call no case drop.

    Limits of this control.

    • The degraded copy was widened after seeing smoke scores (the Fetch channel and Archive-reading sections). A post-hoc amendment, and it still fell short.
    • dd-collision-refuses and mc-resource-seeds-route-away did not run: every Bash-granting case was refused on the Docker store symlink, and the link swap was not used. They name the skill but do not exercise this text, so they would add cases rather than sensitivity. That is my reading, not a measured result.
    • The with-without passes were INVALID (Glob and Grep denials in dd-tagging-rules' without arm), so no VALID baseline arm exists.

    Found. In all three VALID degraded runs the digest kept theme={null} and \& with both profile sections deleted: the verbatim rule in skills/docpage-digest/SKILL.md:205-208 holds quotes byte-exact, not these sentences. Only the reader-link disclosure failed. The profile's Archive-reading bullet also lists the same three artifacts, outside this row.

    Question for the owner: how to settle r318?

  20. removed their assignment
    on Nov 4, 2026
  21. added
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Nov 4, 2026
  22. kyle-sexton commented on Nov 4, 2026

    @kyle-sexton
    MemberAuthor

    Decision recorded. Owner, verbatim, 2026-11-04: "Go w/ recommendations", on the question recommending option (a): keep the sentences and close, with r318 recorded as unmeasurable.

    Row r318 stays as it is in plugins/knowledge/skills/docpage-digest/context/anthropic-docs-profile.md: the negative control fell short (mean drop 0.142 against the 0.15 bar), so under rewrite-a-plugin.md step 7 the before text ships. The other rows were delivered by #12913 and #12818.

  23. removed
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Nov 4, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    host-runNeeds an unattended run on a host; the host runner takes it, not a person or the issue lane.needs-triageNot yet classified. Floor until a type and one priority tier are set.priority: lowNice-to-have, cosmetic, or speculative; opportunistic.work-class: scopedA briefed fix or small feature; blast radius bounded by the brief, tests exist.

    Type

    Fields

    Work status

    agent-ready

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions