The Agent Skills Directory

[COMMAND_EXECUTION]: Executes a local Python script (scripts/render_dossier.py) to process JSON findings into a self-contained HTML report. The script is bundled with the skill and does not have external dependencies.
[EXTERNAL_DOWNLOADS]: Performs web research using external browser automation tools (browser-harness or cursor-ide-browser) to collect client news, technology footprints, and pricing benchmarks.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks when processing untrusted web content or external briefs, though it mitigates some risks via structured data rendering.
Ingestion points: Data is ingested from {run-folder}/brief.md and live web pages during research passes.
Boundary markers: No specific delimiters or "ignore previous instructions" warnings are applied to external content before agent processing.
Capability inventory: The agent can execute local commands (python3, browser-harness) and read/write files in the project workspace.
Sanitization: The render_dossier.py script implements mandatory HTML escaping for all gathered strings to prevent cross-site scripting (XSS) in the final dossier.
[SAFE]: Configuration and findings are stored in localized project paths (e.g., .pawbytes/ and {run-folder}/), ensuring research data is contained and follows vendor-specific organizational patterns.

paw-pa-research