Skip to content

CHANGELOG.md

Latest commit

 

History

History
74 lines (54 loc) · 15.2 KB

CHANGELOG.md

File metadata and controls

74 lines (54 loc) · 15.2 KB
Raw

CHANGELOG — ai-modules

Status markers: [active] — described behavior or code is present and current. [changed later] — still present but meaningfully evolved since this entry was written. [superseded] — replaced, removed, or made obsolete by later work.

Entries are grouped strictly by day and kept on their original implementation dates.

2026-05-10 — git_commit hardening and repo conventions

  • [active] Refactor: Restructured the git_commit skill (v3.2.0) into a phase-based primary workflow with <path_resolution>, <primary_workflow> (gather, read, compose, execute), <fallback_trigger>, <fallback_on_script_failure>, and a consolidated <message_policy> covering multi- and single-file format, line-quality, and a no-attribution-trailers rule, replacing the prior loose <tools> and <steps> layout. Added a sibling references/manual_fallback.md reference describing the only authorized fallback trigger and step-for-step replacements for both bundled scripts plus the post-recovery handoff back to the primary workflow.
  • [active] Refactor/runtime reliability: Hardened prepare_commit_context.sh to route every path list through NUL-delimited git output so paths with newlines or tabs survive intact, cache staged and unstaged numstat output once each so per-file binary detection is O(1), drop the redundant has_diff short-circuit, and emit <staged_new_files> via NUL-safe iteration.
  • [active] Docs/specs-only: Added a Versioning section to CLAUDE.md codifying once-per-session bumps that land in the same commit as the change they cover, with parallel plugin-meta updates across both plugin.json files and the marketplace entry. Added a Regression test harnesses section covering dual-surface testing (bash unit tests for bundled scripts plus skill-creator evals for skill behavior), the per-skill tests/<skill_name>/ layout, the skill-creator-aligned pattern with evals.json, fixtures, and workspace iterations, the legacy tests/wiki/ exception, and the rule that test growth lives in its own session separate from skill-change sessions.
  • [active] Implementation/runtime: Broadened the local-only test-tree exclusion from wiki-tests/ to tests/ in .gitignore, added a matching -path ./tests -prune to the Makefile discovery clause so committed lint never scans the gitignored harness tree, and added tests/ to the CLAUDE.md layout tree.
  • [active] Refactor: Bumped the ai_dev plugin to 1.0.6 across the marketplace entry and both the Claude and Codex plugin manifests.
  • [active] Docs/specs-only: Expanded the root README.md Plugins section with per-skill and per-agent descriptions focused on agent reliability and saved turns and tokens, and added a generic LLM-wiki framing that positions the wiki between a RAG pipeline and loose notes.
  • Files changed: .claude-plugin/marketplace.json, .gitignore, CLAUDE.md, Makefile, README.md, plugins/ai_dev/.claude-plugin/plugin.json, plugins/ai_dev/.codex-plugin/plugin.json, plugins/ai_dev/skills/git_commit/SKILL.md, plugins/ai_dev/skills/git_commit/references/manual_fallback.md, plugins/ai_dev/skills/git_commit/scripts/prepare_commit_context.sh

2026-05-09 — Wiki discovery walk-up and procedure phasing

  • [active] Implementation/runtime: Added walk-up wiki discovery to discover_wiki.sh and lint.py so the wiki skill and wiki_auto_shaper agent climb from the working directory toward the repo root, skip subtrees marked with .no_wiki, and emit a candidate list when the location is ambiguous; updated the agent to prompt the user to select among candidates instead of guessing. Published knowledge_management 1.2.4 across the marketplace and both plugin manifests.
  • [active] Implementation/runtime: Strengthened wiki location-resolution guardrails in the wiki skill and wiki_auto_shaper agent so the agent only proceeds after the user confirms the discovered wiki, with optional .no_wiki marker prompts when discovery is ambiguous. Published knowledge_management 1.2.5.
  • [active] Refactor: Sharpened procedure-versus-concept guidance across the wiki skill, the wiki_auto_shaper agent, CLAUDE.md, and template_schema.md so procedure pages stay focused on operator steps rather than describing the mechanism, and restructured the auto-shaper agent into named phases. Published knowledge_management 1.2.6.
  • [changed later] Refactor: Added wiki-tests/ to .gitignore to keep the local regression-test workspace out of version control.
  • Files changed: .claude-plugin/marketplace.json, .gitignore, CLAUDE.md, plugins/knowledge_management/.claude-plugin/plugin.json, plugins/knowledge_management/.codex-plugin/plugin.json, plugins/knowledge_management/agents/wiki_auto_shaper.md, plugins/knowledge_management/skills/wiki/SKILL.md, plugins/knowledge_management/skills/wiki/references/template_schema.md, plugins/knowledge_management/skills/wiki/scripts/__pycache__/lint.cpython-314.pyc, plugins/knowledge_management/skills/wiki/scripts/discover_wiki.sh, plugins/knowledge_management/skills/wiki/scripts/lint.py

2026-05-06 — Pseudo-XML linter and ai_dev release

  • [active] Implementation/runtime: Expanded the ai_instruction_formatting skill (v2.0.0 → v3.3.0) with a "File Shape" table covering prose-only, XML-instruction body, tutorial with examples, and mixed-agent shapes; an explicit "Rule Strength" split between mechanical (linter-enforced) rules and stylistic guidance; "Named Tags First" and "Repetition as the Exception" guidance with a unit-vs-anchor decision frame and a canonical repeatable-pair allowlist (<policy>/<rule>, <steps>/<step>, <substeps>/<substep>, <examples>/<example>, <scoring_criteria>/<criterion>, <validations>/<validation>); ASCII snake_case tag-name rules with no attributes and no self-closing tags; a depth-five nesting cap; and a "Mechanical Validation" section pointing at the bundled linter.
  • [active] Implementation/runtime: Added scripts/lint_pseudo_xml.py to the ai_instruction_formatting skill — a standalone linter that auto-detects which of the four file shapes the input follows, parses pseudo-XML in SKILL.md and agent files, and emits actionable issues for invalid tag names, attributes, self-closing tags, mismatched and unclosed tags, depth violations, duplicate non-allowlisted siblings, numeric-suffix siblings, frontmatter-vs-directory-name and H1-vs-frontmatter alignment, and trailing-newline hygiene, with info-severity hints at every canonical repetition site so the unit-vs-anchor judgment is made explicitly each time.
  • [active] Refactor: Replaced the update_changelog skill's <steps> block of <step name="..."> attribute-bearing entries with a <procedure> block of self-describing tags (<resolve_project_name>, <enumerate_dates>, <ensure_header>, <day_loop> with nested <substeps>, <status_re_evaluation>), removing every XML attribute from the file.
  • [changed later] Refactor: Bumped the ai_dev plugin to 1.0.5 across .claude-plugin/marketplace.json, the Claude plugin manifest, and the Codex plugin manifest.
  • Files changed: .claude-plugin/marketplace.json, plugins/ai_dev/.claude-plugin/plugin.json, plugins/ai_dev/.codex-plugin/plugin.json, plugins/ai_dev/skills/ai_instruction_formatting/SKILL.md, plugins/ai_dev/skills/ai_instruction_formatting/scripts/lint_pseudo_xml.py, plugins/ai_dev/skills/update_changelog/SKILL.md

2026-05-05 — Changelog tooling and skill rewrite

  • [active] Implementation/runtime: Bootstrapped CHANGELOG.md at the repository root with the day-grouped status-marker legend and initial entries for the 2026-05-03 bootstrap and 2026-05-04 wiki-audit-hardening days.
  • [changed later] Refactor: Rewrote the update_changelog skill (v2.0.0 → v3.0.0) from prose markdown into a pseudo-XML <update_changelog_skill> body with explicit <objective>, <tools>, <output_contract>, <policy>, and <steps> sections, and pointed the workflow at the bundled prepare_changelog_day.sh helper.
  • [active] Implementation/runtime: Added scripts/prepare_changelog_day.sh to the update_changelog skill — a single-call helper that emits one structured <changelog_day> blob per date with commit subjects and bodies, the deduplicated repo-relative file list, per-file net diffs across the day's first-parent-to-last range, generic placeholders for binary files, and an empty-tree fallback for the repo's root commit.
  • [superseded] Refactor: Bumped the ai_dev plugin from 1.0.1 to 1.0.2 across the Claude and Codex manifests.
  • Files changed: CHANGELOG.md, plugins/ai_dev/.claude-plugin/plugin.json, plugins/ai_dev/.codex-plugin/plugin.json, plugins/ai_dev/skills/update_changelog/SKILL.md, plugins/ai_dev/skills/update_changelog/scripts/prepare_changelog_day.sh

2026-05-04 — Wiki audit hardening and attribution

  • [active] Docs/specs-only: Rewrote the README intro to position the Claude marketplace, global make deploy, project-scoped --project-dir, and in-place checkout as equal distribution paths instead of primary plus fallback.
  • [active] Docs/specs-only: Added the deployment-agnostic cross-references authoring convention to CLAUDE.md so artefacts name siblings directly (wiki_auto_shaper, format_markdown) without qualifying by plugin, marketplace, or deployment path.
  • [active] Implementation/runtime: Broadened the wiki skill trigger phrasing to include fix, clean-up, and auto-repair, renamed its lint section to "Lint and Audit", split guidance into broad audits delegated to the wiki_auto_shaper agent versus narrow inline lint.py runs after a single ingest or archive, and extended the agent trigger to match "fix" requests.
  • [active] Implementation/runtime: Added canonical scaffold-drift checks to the wiki_auto_shaper agent for SCHEMA.md, index.md, log.md, the directory layout, and raw-source frontmatter, then retooled them to drive comparisons from a mechanical diff -u against canonical templates with preserve/drift hunk classification and configurable authoritative zones (Domain body, Tag Taxonomy, custom frontmatter, custom page types, index headers and entries, log entries).
  • [active] Implementation/runtime: Added the boilerplate lint check (check_verbatim_boilerplate with a VerbatimSlot dataclass and a default extract_h1_prelude extractor) wired into lint.py main(), propagated the canonical "managed by the wiki skill" attribution paragraph through template_schema.md between H1 and ## Domain, and replaced the Karpathy attribution paragraph in the wiki skill with the lint-enforced verbatim variant.
  • [active] Docs/specs-only: Added a Background section to the knowledge_management plugin README crediting Karpathy's LLM Wiki pattern and explaining what the wiki skill keeps (compile-once-and-compound), ports (tool-neutral plain markdown), and adds (procedure pages, scripts, the wiki_auto_shaper agent, sha256 provenance, schema-driven page-type enum).
  • [changed later] Refactor: Bumped plugin and marketplace versions — knowledge_management from 1.0.0 to 1.2.3 across the marketplace and both plugin manifests, ai_dev from 1.0.0 to 1.0.1, the wiki skill from 1.5.x to 1.7.0, and the wiki_auto_shaper agent to 1.3.2.
  • Files changed: .claude-plugin/marketplace.json, CLAUDE.md, README.md, plugins/ai_dev/.claude-plugin/plugin.json, plugins/ai_dev/.codex-plugin/plugin.json, plugins/knowledge_management/.claude-plugin/plugin.json, plugins/knowledge_management/.codex-plugin/plugin.json, plugins/knowledge_management/README.md, plugins/knowledge_management/agents/wiki_auto_shaper.md, plugins/knowledge_management/skills/wiki/SKILL.md, plugins/knowledge_management/skills/wiki/references/lint_checks.md, plugins/knowledge_management/skills/wiki/references/template_schema.md, plugins/knowledge_management/skills/wiki/scripts/lint.py

2026-05-03 — Repo bootstrap and plugin scaffolding

  • [active] Implementation/runtime: Bootstrapped the repository with the knowledge_management plugin (wiki skill plus references and scripts), the multi-vendor deployment.sh discovery script, the Claude marketplace manifest under .claude-plugin/, and MIT licensing.
  • [active] Implementation/runtime: Expanded the knowledge_management plugin with the executive_summary and spr skills.
  • [changed later] Docs/specs-only: Clarified repository and plugin descriptions to position the modules for multi-vendor deployment across Claude Code, Codex, Cursor, Copilot, Gemini CLI, and Antigravity.
  • [active] Implementation/runtime: Added a Make entry point with deploy, uninstall, lint, fix, and help targets, alongside .markdownlint.jsonc configured for prompt-oriented markdown (MD033 off for pseudo-XML, MD013/MD041 off).
  • [active] Implementation/runtime: Added the ai_dev plugin with the ai_instruction_writing, ai_instruction_formatting, format_markdown, format_python, format_rust, git_commit, and update_changelog skills.
  • [changed later] Docs/specs-only: Added CLAUDE.md operating guidance covering layout, common Make tasks, skill and plugin editing workflow, and authoring conventions for AI agents working in this meta-repository.
  • [changed later] Refactor: Moved the git_commit skill to script-backed context gathering and execution via prepare_commit_context.sh and commit_with_message.sh.
  • [changed later] Implementation/runtime: Added the wiki_auto_shaper agent for assess-fix-verify wiki audits (initially named wiki_audit, renamed the same day) and registered it in the knowledge_management plugin metadata, marketplace entry, and READMEs.
  • Files changed: .claude-plugin/marketplace.json, .gitignore, .markdownlint.jsonc, CLAUDE.md, LICENSE, Makefile, README.md, deployment/README.md, deployment/deployment.conf, deployment/deployment.sh, plugins/ai_dev/.claude-plugin/plugin.json, plugins/ai_dev/.codex-plugin/plugin.json, plugins/ai_dev/README.md, plugins/ai_dev/skills/ai_instruction_formatting/SKILL.md, plugins/ai_dev/skills/ai_instruction_writing/SKILL.md, plugins/ai_dev/skills/format_markdown/SKILL.md, plugins/ai_dev/skills/format_python/SKILL.md, plugins/ai_dev/skills/format_rust/SKILL.md, plugins/ai_dev/skills/git_commit/SKILL.md, plugins/ai_dev/skills/git_commit/scripts/commit_with_message.sh, plugins/ai_dev/skills/git_commit/scripts/prepare_commit_context.sh, plugins/ai_dev/skills/update_changelog/SKILL.md, plugins/knowledge_management/.claude-plugin/plugin.json, plugins/knowledge_management/.codex-plugin/plugin.json, plugins/knowledge_management/README.md, plugins/knowledge_management/agents/wiki_auto_shaper.md, plugins/knowledge_management/skills/executive_summary/SKILL.md, plugins/knowledge_management/skills/spr/SKILL.md, plugins/knowledge_management/skills/wiki/SKILL.md, plugins/knowledge_management/skills/wiki/references/lint_checks.md, plugins/knowledge_management/skills/wiki/references/template_index.md, plugins/knowledge_management/skills/wiki/references/template_log.md, plugins/knowledge_management/skills/wiki/references/template_schema.md, plugins/knowledge_management/skills/wiki/scripts/discover_wiki.sh, plugins/knowledge_management/skills/wiki/scripts/init_wiki.sh, plugins/knowledge_management/skills/wiki/scripts/lint.py