The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run with list-based arguments in scripts/github_clone.py to manage local repository clones using git and rm. This is a standard and safe implementation for its intended purpose of repository management.
[EXTERNAL_DOWNLOADS]: The skill fetches content from several well-known and trusted services including GitHub (via git clone), Readwise (via readwise CLI), NotebookLM (via nlm CLI), and the web (via brightdata-cli). These operations are consistent with the skill's primary purpose and target official service endpoints.
[EXTERNAL_DOWNLOADS]: The scripts/download_assets.py script uses the httpx library to download remote images referenced in markdown files. This is used to make research artifacts self-contained and is limited to image file types.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection (Category 8) because it ingests untrusted data from various external sources (web, PDF, GitHub) which are then processed by subagents to generate synthesis and wiki pages.
Ingestion points: Step 1 in SKILL.md identifies ingestion of brain dumps, vault paths, web URLs, Readwise highlights, NotebookLM notes, GitHub repositories, and local/web PDFs.
Boundary markers: Explicit delimiters for untrusted content are not defined in the subagent prompts, though tasks are highly structured (e.g., 'Identify 2-6 key claims').
Capability inventory: The skill can execute shell commands via scripts/github_clone.py (git, rm), perform file operations in agents/builder.md (cp, printf, jq), and make network requests in scripts/download_assets.py (httpx).
Sanitization: Filenames are sanitized using slugification (kebab-case, ASCII only, max 60 chars) as described in Step 6 of SKILL.md and CONVENTIONS.md.

research