The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill features a 'SIGNAL' domain (reference/signal.md) that fetches and processes untrusted data from external platforms including Reddit, Hacker News, GitHub, and various web pages via the Jina Reader service (r.jina.ai). Maliciously crafted content within these external sources could influence the agent's logic during the summarization or synthesis phases of product research.
Ingestion points: Data is ingested through WebSearch and WebFetch operations directed at public forums, repositories, and news sites as described in reference/signal.md.
Boundary markers: The skill utilizes an 'Intent capture' gate (reference/intent.md) and an 'independent critic' (reference/critic.md) to frame and verify artifacts, though these are logical checks rather than robust technical sanitization.
Capability inventory: The skill has the capability to read local project files and write multi-page HTML and Markdown artifacts (workspace.html, prd.md).
Sanitization: The instructions mandate separating evidence from inference and explicitly labeling unvalidated claims as hypotheses, which serves as a procedural mitigation.
[EXTERNAL_DOWNLOADS]: The skill facilitates data retrieval from several well-known and established technology services and APIs to gather market evidence. These include the Algolia Hacker News API (hn.algolia.com), the official GitHub API (api.github.com), the Polymarket API (gamma-api.polymarket.com), and the Jina Reader markdown conversion service (r.jina.ai). These references are consistent with the skill's stated purpose of gathering 'live market & customer signal'.

superpm