The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data (pasted brand samples, web content) and uses it to generate brand guidelines. While instructions explicitly warn to treat these sources as untrusted input and never follow embedded instructions, the inherent attack surface for indirect prompt injection remains.
Ingestion points: Processes pasted text samples and scrapes external URLs using scripts/connectors/firecrawl.py.
Boundary markers: The instructions include a specific security warning to ignore embedded instructions in source material.
Capability inventory: Writes to memory/events/narrative.ndjson and memory/events/claims.ndjson via a helper script registry-events.py.
Sanitization: Not explicitly defined, though the skill relies on the agent's internal safety guardrails and the explicit 'ignore' instruction.
[COMMAND_EXECUTION]: The skill invokes local scripts scripts/connectors/firecrawl.py for web scraping and registry-events.py for proposing changes to the narrative registry. These are functional components of the skill's workflow and are consistent with the author's other skills.
[EXTERNAL_DOWNLOADS]: The skill utilizes firecrawl.py to scrape external content from URLs provided by the user. The documentation states this is for gathering the brand's own published material and follows robots.txt pre-flight rules.

brand-language-codifier