[abullard1]

Fixes #7. Thanks to @aermak for the report and for identifying the cause.

On native Windows the plugin was silently inert. No bolding, no error, and /claude-bionify:status reporting ON throughout.

Cause

hooks.json declared one interpreter, python3. Hooks run in exec form, which Claude Code spawns with no shell, so the name must resolve to a real executable. On native Windows it usually does not:

• The python.org installer creates python.exe and py.exe, never a python3.exe.
• Windows ships a 0-byte App Execution Alias stub named python3.exe in WindowsApps, on PATH and enabled by default. Running it prints Python was not found and exits 9009.

So the spawn succeeds and runs the wrong thing. Existence checks like where python3 pass while the plugin stays dead. A failed MessageDisplay hook is display-only, so Claude Code keeps the original delta and surfaces nothing; claude --debug is the only trace.

Fix

{"command": "python3", "args": ["${CLAUDE_PLUGIN_ROOT}/scripts/bionify.py"], "timeout": 10},
{"command": "py",      "args": ["-3", "${CLAUDE_PLUGIN_ROOT}/scripts/bionify.py"], "timeout": 10}

Entries dedupe on command plus args so both are attempted, a failing one yields no displayContent, and the merge keeps the last defined one. python3 covers Linux, macOS, WSL and Store Python; py covers python.org on Windows, whose launcher is a real .exe in C:\Windows.

Not chosen: a python third entry, which on Arch and in any activated venv is the same interpreter as python3 and would run the transform twice per flush; and shell form with a || chain, which fails to parse under PowerShell 5.1, the fallback when Git Bash is absent.

/claude-bionify:status also stopped reporting ON while nothing was bolded, which is what made this look like a settings problem.

Known limitation

Where python3 is the Store placeholder, bolding works but the /claude-bionify:* commands do not, since they run through a shell. Installing Git for Windows, or a real python3 on PATH, restores them. Documented in both READMEs.

Test plan

• 115 passed, 2 skipped, 93.7% coverage (gate 80%)
• Each fix fails a named test when reverted
• hooks.json contract asserted in the suite, since claude plugin validate --strict does not check it
• Spawn test runs the hook under each declared interpreter by name
• claude plugin validate --strict on both manifests, ruff clean

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.