{ "intent": "Clear every over-cap markdown file in the repo, splitting what is still reference and deleting what is a record of finished work.", "scope": [ "Split: linux_configuration/docs/SECURITY_HARDENING_ANALYSIS.md (761 -> 16 + 6 parts)", "Split: docs/superpowers/specs/2026-05-01-phone-focus-recovery-design.md (709 -> 51 + 5 parts)", "Split: .github/skills/test-driven-development/SKILL.md (384 -> 135) and code-review-and-quality/SKILL.md (358 -> 123), detail into references/", "Split: pacman/README_FOR_LLM.md (338), phone_focus_mode/README.md (336), README_MIDNIGHT_SHUTDOWN_LLM.md (300), PACMAN_WRAPPER_SECURITY.md (258)", "Deleted: docs/cleanup-2026-05/ (7 files, 1524 lines) and linux_configuration/docs/VERIFICATION.md (261)", "Non-goal: no prose rewritten or condensed" ], "changes": [ "Eight files split into per-topic subdirectories, each original left as a short index linking to its parts. Both SKILL.md files keep their frontmatter and description so skill discovery is unaffected; only the detail sections moved to references/.", "docs/cleanup-2026-05/ deleted on the user's explicit instruction. Its seven files are status reports of a finished 2026-05 cleanup ('What Was Delivered', 'After Your Changes'), nothing references them, and the rules they describe are now enforced by the no-polling-antipatterns, shellcheck and shfmt hooks and documented in the efficient-polling-scripts skill. The only forward-looking content was a speculative 'Backlog / Future Work' list never actioned in fifteen months.", "linux_configuration/docs/VERIFICATION.md deleted as the same class: an unreferenced '✅ Implementation: COMPLETE' checklist of work already done. Applied the user's ruling consistently rather than splitting a file of identical character.", "PACMAN_WRAPPER_SECURITY.md was split rather than deleted - unlike VERIFICATION.md it documents live security guarantees and is genuine reference." ], "verification": [ { "command": "bash ~/utils/scripts/check_file_length.sh --all | grep '\\.md:'", "result": "pass", "evidence": "No output. Every markdown file in the repo is now under the cap, down from twelve violations." }, { "command": "diff <(git show HEAD: | grep -vE '^\\s*$' | sort) <(cat | grep -vE '^\\s*$' | sort) | grep -c '^<'", "result": "pass", "evidence": "0 lines lost for both SKILL.md splits, checked explicitly. The two largest splits were verified by md5 of their sorted non-blank lines: a6c0f601 for SECURITY_HARDENING_ANALYSIS and b5b03a5e for the phone-focus spec, identical on both sides." }, { "command": "grep -rn 'VERIFICATION.md|cleanup-2026-05' over all md/yaml/json/sh", "result": "pass", "evidence": "No hits outside append-only evidence and contract artifacts, so the deletions leave no dangling link. copilot-instructions.md references SECURITY_HARDENING_ANALYSIS.md, which was split rather than deleted and still resolves." }, { "command": "pre-commit run --files <38 changed files>", "result": "pass", "evidence": "Clean, including prettier and the markdown hooks." } ], "risks": [ "Deleting 1785 lines is the least reversible change in this whole effort. It rests on the user's explicit instruction plus three checks: nothing references the files, their rules are hook-enforced, and their content is status reporting rather than reference. Git history retains them.", "The SKILL.md files are now split across SKILL.md and references/. If the skill loader only reads SKILL.md, the detail is no longer auto-loaded - it is linked, not inlined. That is the documented references/ pattern, but it does change what an agent sees without following a link." ], "rollback": [ "git revert the commit; the eight splits recombine and both deleted paths return.", "After rollback, re-run check_file_length.sh and expect twelve markdown violations again." ] }