The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a shell command to execute a local Python script provided within the skill package.
Evidence: SKILL.md (Step 7) instructs the agent to run python3 "${CLAUDE_SKILL_DIR}/scripts/build_pack.py" workflow-packs/<slug>/pack.md to generate an HTML preview.
[DYNAMIC_EXECUTION]: The skill follows a pattern of generating content (markdown) and then executing a local script to process and transform that content at runtime.
Evidence: The workflow involves writing to pack.md and immediately calling the build_pack.py script to generate pack.html.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input to generate executable prompts and HTML files, creating an attack surface where malicious input could influence the generated output.
Ingestion points: The skill collects the user's "idea or reference" and "end output" description in SKILL.md (Step 1).
Boundary markers: Instructions use template placeholders (e.g., {{STEP_1_PROMPT}} in references/pack-template.md) to structure the generated content.
Capability inventory: The skill performs file writes (pack.md, pack.html) and executes a Python script via a subprocess call.
Sanitization: The provided conversion script (scripts/build_pack.py) uses html.escape() to sanitize content before placing it in the generated HTML page.

workflow-maker