claudemd-cli 0.48.0 → 0.49.0

package/CHANGELOG.md

@@ -8,6 +8,26 @@ All notable changes to the `claudemd` plugin. This changelog tracks plugin artif
 - **Canonical spec version source**: `spec/CLAUDE.md` top-line title (`# AI-CODING-SPEC vX.Y.Z — Core`) + `spec/CLAUDE-changelog.md` top `##` entry.
 - **Plugin semver vs spec semver** are independent: plugin patch (0.2.0 → 0.2.1) may ship when spec is unchanged (this release); plugin minor (0.1.9 → 0.2.0) ships when spec minor updates (v0.2.0 shipped spec v6.10.0).
 
+## [0.49.0] - 2026-07-15
+
+Spec **v6.20.0** — §2.1 Model tiering rule removed (spec-only; no hook/script change).
+
+- **Removed the spawned-agent model-tiering rule** (core §2.1 paragraph + `spec/CLAUDE-extended.md §2.1-EXT MODEL TIERING`): downgrade-eligible category enumeration, the NEVER-downgrade list, the verifier ≥ generator invariant, and the anomalous-output re-run clause are all deleted. The model now self-allocates subagent tiers on its own judgment — quality-first with zero spec constraint. Rationale: the `Agent` tool already defaults to inheriting the parent (session) model when `model` is omitted, so quality-first is the harness default without any spec text. SHOULD-level rule — `spec/hard-rules.json` untouched, so no hook behavior changes and no enforcement surface moves.
+- **Why now**: the first real-world sample of the rule firing (2026-07-15, a sibling project running subagents) showed the orchestrator self-allocating verify/review subagents to a lower tier including same-tier self-review — the exact pattern the guardrail targeted. The operator chose to trust the model's own allocation over re-adding the constraint (tracked in the durable-memory note `feedback_tiering_verify_downgrade_gap.md`, reopenable if the pattern recurs and hurts quality).
+- **Manifest descriptions** bumped `AI-CODING-SPEC v6.19 → v6.20` (major.minor per the versioning policy above). README spec-version references bumped to v6.20.
+- **Tests**: full `npm test` green; `node scripts/version-cascade-check.js` exit 0 (v6.20 consistent across spec trio + README + 3 manifests; plugin semver 0.49.0 consistent across 4 sites; Sizing drift within ±20B).
+
+## [0.48.1] - 2026-07-15
+
+**Patch — three §8 pre-detector silent bypasses closed (D1 brace-subpath, D2 arithmetic/quoted `<<` fakes a heredoc). Bugfix restoring intended deny behavior; no new gate, no §8 detector added.**
+
+- **D1 — `canon_cmd_words` tore `${VAR}/subpath` apart.** It set command position on `{`, so `rm -rf "${SP}/build"` was basenamed to `rm -rf "${build"`, erasing the `${SP}` the var-detector greps for → the segment was skipped (ALLOW). Only shapes with a `/` after the `{` were mangled, landing the bypass precisely on the dangerous subpath class; `rm -rf "${HOME}/"` (empty `HOME` = `rm -rf /`, the ValveSoftware/steam-for-linux#3671 residue case) bypassed too. Fix: `{` re-opens command position only as a real brace-group introducer (`{ rm; }`), guarded on the preceding char not being `$`. Introduced with `canon_cmd_words` in v0.42.0 (SEC-2).
+- **D2 — a `<<` left-shift / quoted `<<` faked a heredoc and blanked the following command.** `heredoc_re` matched any `<<word`, so `$((1<<bits))`, `$[a<<b]`, `let a<<b`, and a `<<` inside a quoted string (`echo "a<<b"; rm -rf $EVIL`) all set `in_heredoc`, truncated the line at `<<`, and blanked every following line — deleting the rm/npx/curl from the text **all three** detectors scan (blinds the whole gate, not one detector). Fix (root cause, closes the class without enumerating shell syntaxes): treat `<<TAG` as a heredoc only when a matching terminator line actually exists later. A genuine heredoc always closes with its tag; a shift / quoted `<<` / comparison never does. The new heredoc-detection condition is a strict `AND`-narrowing of the old one, so it can only make detection *less* aggressive → expose more text to the deny-on-match detectors → never a new bypass (proven by a differential old-vs-new sweep: 0 inputs newly allowed).
+- **Found by** an adversarial fresh-subagent review (2026-07-15). D1/D2 were reproduced against the shipped hook, fixed, and RED-proven; the review's follow-up (`$[…]` / `let` / quoted `<<`, all the same class) folded into the same terminator-guard fix and a second review round confirmed the invariant holds.
+- **Also — multi-window refresh note.** `/claudemd-refresh`, `refresh-plugin.sh`, and README §Update now tell users to run `/reload-plugins` in **other open Claude Code windows**: a refresh removes the old versioned plugin-cache dir those sessions pinned their hook paths to at startup, so they error on every hook event (enforcement absent) until reloaded.
+- **Residuals** (documented in the hook; deliberate-crafting territory that does not clear the "ordinary mistake" bar, so not chased): indirect-name rebind (`unset "$T"`, `trap 'S=' DEBUG`), and a fake heredoc whose tag is repeated as a bare line to acquire a coincidental terminator (present in the original code too — this fix does not widen it). `IFS=/; rm -rf $SP/build` was probed and denies.
+- **Tests**: pre-bash-safety corpus +17 rows (10 deny + 7 FP-pass controls) in `tests/fixtures/bash-safety/corpus.tsv`, all RED-proven against the committed pre-fix hook (347→357 with the fix). Full `npm test` green.
+
 ## [0.48.0] - 2026-07-15
 
 **Minor — one-command plugin refresh: `/claudemd-refresh`.** Closes the update-UX gap: the upgrade banner used to teach a 4-command paste sequence; now it names one command. Detection (v0.4.0 `upstream_check`, 24h-throttled `git ls-remote`) and post-refresh spec/manifest sync (version-sync hook / SessionStart bootstrap) were already automatic — this release ships the missing middle step.

package/README.md

@@ -1,12 +1,12 @@
 # claudemd
 
-> A **personal AI-coding discipline harness**: one developer's opinionated **AI-CODING-SPEC v6.19**, encoded as Claude Code shell hooks and shipped with the plugin. Built and dogfooded on my own repos — fork and adapt, don't adopt wholesale.
+> A **personal AI-coding discipline harness**: one developer's opinionated **AI-CODING-SPEC v6.20**, encoded as Claude Code shell hooks and shipped with the plugin. Built and dogfooded on my own repos — fork and adapt, don't adopt wholesale.
 
 [![CI](https://github.com/sdsrss/claudemd/actions/workflows/ci.yml/badge.svg?branch=main)](https://github.com/sdsrss/claudemd/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/claudemd-cli.svg)](https://www.npmjs.com/package/claudemd-cli)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 
-claudemd plugs into the Claude Code hook system to **block commits, pushes, and bash commands** that violate AI-CODING-SPEC v6.19 — banned vocabulary in commit messages, `rm -rf $VAR` without variable validation, ship-on-red-CI, unread `MEMORY.md` entries during release flows, and more. The spec itself (`CLAUDE.md` + `CLAUDE-extended.md` + `CLAUDE-changelog.md` + `OPERATOR.md`) ships with the plugin and installs into `~/.claude/`, so the rules Claude Code reads at session start match the rules the hooks enforce. (`OPERATOR.md` is the human-only spec-maintenance handbook — Agent-loaded files are the CLAUDE trio.)
+claudemd plugs into the Claude Code hook system to **block commits, pushes, and bash commands** that violate AI-CODING-SPEC v6.20 — banned vocabulary in commit messages, `rm -rf $VAR` without variable validation, ship-on-red-CI, unread `MEMORY.md` entries during release flows, and more. The spec itself (`CLAUDE.md` + `CLAUDE-extended.md` + `CLAUDE-changelog.md` + `OPERATOR.md`) ships with the plugin and installs into `~/.claude/`, so the rules Claude Code reads at session start match the rules the hooks enforce. (`OPERATOR.md` is the human-only spec-maintenance handbook — Agent-loaded files are the CLAUDE trio.)
 
 A standalone CLI (`npx claudemd-cli`) reuses the same `banned-vocab.patterns` source for git pre-commit hooks, GitHub Actions, and other agents that don't run inside Claude Code.
 
@@ -60,7 +60,7 @@ Verify in one command (Linux): `node --version && jq --version && gh --version &
 | 16 shell hooks | `banned-vocab-check` · `pre-bash-safety-check` · `ship-baseline-check` · `residue-audit` · `memory-read-check` · `memory-prompt-hint` · `mid-spine-yield-scan` · `sandbox-disposal-check` · `session-start-check` · `session-extended-read` · `session-summary` · `session-end-check` · `transcript-vocab-scan` · `transcript-structure-scan` · `version-sync` · `mem-audit` |
 | 16 slash commands | `/claudemd-install` · `/claudemd-status` · `/claudemd-update` · `/claudemd-refresh` · `/claudemd-audit` · `/claudemd-toggle` · `/claudemd-doctor` · `/claudemd-analyze` · `/claudemd-uninstall` · `/claudemd-rules` · `/claudemd-clean-residue` · `/claudemd-sparkline` · `/claudemd-sampling-audit` · `/claudemd-bypass-audit` · `/claudemd-design-adopt` · `/claudemd-statusline` |
 | 1 standalone CLI | `claudemd-cli lint` · `claudemd-cli audit` ([npm: `claudemd-cli`](https://www.npmjs.com/package/claudemd-cli)) |
-| Spec v6.19 | `~/.claude/CLAUDE.md` · `CLAUDE-extended.md` · `CLAUDE-changelog.md` · `OPERATOR.md` (backup-before-overwrite) |
+| Spec v6.20 | `~/.claude/CLAUDE.md` · `CLAUDE-extended.md` · `CLAUDE-changelog.md` · `OPERATOR.md` (backup-before-overwrite) |
 | StatusLine (opt-out) | PS1-style line — `user@host:dir (branch) Model [ctx:N% · 5h:N% · 7d:N%]` (`dir` = cwd basename; context / 5-hour quota / weekly quota, all **used %**, read from Claude Code's `rate_limits` payload; quota segments auto-hide when the data is absent, or force-hide with `DISABLE_STATUSLINE_QUOTA=1`) — wired into `~/.claude/settings.json` on install **only when the slot is empty**; an existing statusline is left untouched. Skip entirely with `CLAUDEMD_NO_STATUSLINE=1`. Manage via `/claudemd-statusline`. |
 
 Install backs up a hand-written `~/.claude/CLAUDE.md` (any file without the `# AI-CODING-SPEC` H1) to `~/.claude/backup-<ISO>/` before overwriting (last 5 kept automatically). An already-installed claudemd spec is overwritten **without** a backup — deliberate (v0.23.11): the sole backup is always your own content, so `restore` can never return a stale spec instead. Uninstall offers `keep / restore / delete`; `delete` requires an extra confirmation.
@@ -260,6 +260,8 @@ Manual fallback (e.g. the `claude` CLI is not on PATH):
 
 Or open the interactive UI via `/plugin` → **Installed** tab → select `claudemd` → follow upgrade prompts.
 
+> **Other open Claude Code windows:** run `/reload-plugins` in each of them too. A refresh removes the old versioned plugin-cache dir, but every already-running session pinned its hook paths to that dir at startup — those windows error on every hook event (claudemd enforcement is off there) until they reload or restart.
+
 After the plugin upgrade, sync the shipped spec into `~/.claude/`:
 
 ```
@@ -345,7 +347,7 @@ claudemd/
 ├── commands/                 # 16 slash-command markdown files
 ├── bin/                      # standalone CLI entrypoint (claudemd-lint.js → `npx claudemd-cli` on npmjs.org)
 ├── scripts/                  # 18 Node.js scripts + scripts/lib/ (single-source registry, lint, etc.)
-├── spec/                     # shipped v6.19.0 CLAUDE*.md trio + OPERATOR.md + hard-rules.json manifest
+├── spec/                     # shipped v6.20.0 CLAUDE*.md trio + OPERATOR.md + hard-rules.json manifest
 ├── tests/                    # hook shell tests + Node.js tests + integration + fixtures
 ├── docs/                     # ADDING-NEW-HOOK.md + RULE-HITS-SCHEMA.md + superpowers/
 └── .github/workflows/        # ci.yml (ubuntu+macOS × node 20) + npm-publish.yml (tag-triggered)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudemd-cli",
-  "version": "0.48.0",
+  "version": "0.49.0",
   "description": "Standalone CLI for §10-V banned-vocab + transcript scanning. Companion to the claudemd Claude Code plugin (github.com/sdsrss/claudemd) for use in git pre-commit hooks, GitHub Actions, and other agents.",
   "type": "module",
   "bin": {