The Agent Skills Directory

[SAFE]: The bundle manages API credentials securely through environment variables and provides .env.example templates to prevent hardcoding of secrets.
[SAFE]: Included utility scripts like scripts/check_no_secrets.py and scripts/sync_codex_marketplace.py verify that sensitive environment files or personal data are not accidentally tracked or published in version control.
[INDIRECT_PROMPT_INJECTION]: The skills ingest untrusted content from Facebook posts and comments. Instruction files contain specific guidance directing the agent to treat this data as information rather than instructions. 1. Ingestion points: Pasted post text and comments in fb-hook-extractor and fb-engagement-drafter. 2. Boundary markers: Explicit instructions in skills/fb-humanizer/sub-skills/story-bank-interview.md warn against following embedded commands. 3. Capability inventory: Network API calls and local file system writes to profile files. 4. Sanitization: Instruction-level separation of data and instruction context.
[COMMAND_EXECUTION]: A 'Tier 2' feature in lib/backend_selector.py enables the execution of a user-specified CLI tool via the FB_SKILLS_CUSTOM_POSTER environment variable. While this is an intentional integration point for advanced users, it involves dynamic process execution. The risk is minimized by passing data via JSON on standard input.
[EXTERNAL_DOWNLOADS]: The requirements-lock.txt file references package versions (e.g., requests 2.34.2 and certifi 2026.7.22) that do not currently exist in public registries. While likely intended to match the author's futuristic '2026' theme, these unverifiable dependencies represent a potential supply-chain surface if not properly validated.

facebook-marketing