The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill bundle contains a significant surface for processing untrusted external content.
Ingestion points: The ig-hook-extractor and ig-repurposer skills take user-supplied captions and source text, while ig-audience-insights fetches data from Apify actors (hashtag and profile scrapers).
Boundary markers: The skill instructions rely on Natural Language boundaries but lack explicit programmatic delimiters or "ignore instructions" system warnings when processing this untrusted data.
Capability inventory: The skill possesses capabilities for network operations (requests), local file reading (for media uploads), and command execution (via the DIY poster mode).
Sanitization: While the ig-humanizer component cleans text for stylistic "AI tells," it does not perform security-centric sanitization to prevent the agent from obeying instructions embedded in the ingested marketing data.
[DYNAMIC_EXECUTION]: The lib/backend_selector.py component supports an advanced "DIY" publishing mode.
It utilizes subprocess.run to execute a shell command defined in the INSTAGRAM_SKILLS_CUSTOM_POSTER environment variable.
Although it uses shlex.split for argument parsing and list-based execution (mitigating basic shell injection), it allows the agent to trigger arbitrary user-configured local binaries on approval.
[EXTERNAL_DOWNLOADS]: The skill performs legitimate network operations to well-known service providers to support its core functions.
It fetches niche performance data from the Apify API (api.apify.com).
It uploads media and schedules content through the Publora API (api.publora.com).
It generates on-brand illustrations using the Pixfaro API (api.pixfaro.com).

instagram-marketing