The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/save_session_resume.py uses subprocess.run to interact with the macOS pbcopy and pbpaste utilities. These calls are implemented securely using argument lists (e.g., ["pbcopy"]) instead of shell strings, which prevents shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes session transcripts and transcript metadata, which are potentially untrusted data sources.
Ingestion points: Data is ingested via the session-discovery skill and transcript metadata as described in SKILL.md.
Boundary markers: The agent is instructed to shell-quote directory paths and wrap the generated resume command in Markdown code fences.
Capability inventory: The skill possesses capabilities for local file writing (index.write_text) and system clipboard modification (pbcopy). It lacks network access or direct execution capabilities.
Sanitization: The Python script save_session_resume.py enforces that input fields are single lines and explicitly forbids the command argument from containing Markdown code fences (""" in command`), preventing schema confusion in the resulting Markdown file.

save-session-resume