decorate

/decorate

Manage the prompt-decorators catalogue, always-on list, and auto-decorate mode. The actual work happens in a Python dispatcher; this command is a thin shell around it.

The user invoked: /decorate $ARGUMENTS

Contract

GOAL: Show the user the exact, unmodified output of the dispatcher so they can act on it (e.g., see the decorator list, copy a sigil, confirm an auto mode change). Testable: the output in the fenced block matches what scripts/dispatch.py prints verbatim.

CONSTRAINTS:

• Command is read-mostly. The only writes are to ~/.config/prompt-decorators/config.json via enable/disable/always/auto subcommands - never to project files.
• $ARGUMENTS must reach the dispatcher as tokenised argv, NOT as a shell command string. The heredoc below uses a quoted delimiter ('PD_ARGS_EOF') so user-supplied metacharacters cannot escape; the dispatcher then uses shlex.split inside Python for safe tokenisation.
• Do not interpret or summarise the output unless the user explicitly asks. Relay verbatim.

FORMAT: A fenced code block containing the dispatcher's stdout. If the dispatcher exits non-zero, include the exit code and any stderr.

FAILURE CONDITIONS:

• Output is paraphrased rather than verbatim.
• $ARGUMENTS is interpolated into a shell command string (reintroduces the injection vector).
• The user's config.json is written outside the subcommand's intent (e.g., running list leaves the config modified).

Process

1. Invoke the dispatcher via a quoted heredoc so shell metacharacters in the user's input can't be executed:

   python3 "${CLAUDE_PLUGIN_ROOT}/scripts/dispatch.py" --args-from-stdin <<'PD_ARGS_EOF'
   $ARGUMENTS
   PD_ARGS_EOF
   

2. Wrap the output in a fenced code block and return it to the user without edits.

3. If the user's query is "what decorator should I use for X?" rather than a subcommand, invoke the prompt-decorators-usage skill instead of running this dispatcher - it's better suited to that kind of judgement call.