The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted data from the internet to perform research.
Ingestion points: Web content retrieved via WebFetch and WebSearch in SKILL.md.
Boundary markers: Absent; there are no specific delimiters used to separate untrusted web content from the agent's instructions.
Capability inventory: The skill utilizes Write to save files locally and Task to delegate work to other agents.
Sanitization: The skill includes a mandatory Phase 3 verification step where factual claims must be cross-checked against sources.
[EXTERNAL_DOWNLOADS]: The skill is configured to perform up to 5 web searches and 5 file reads via the chronicle-researcher sub-agent to gather academic and journalistic material.
[COMMAND_EXECUTION]: The skill uses the Task tool to programmatically delegate the research phase to a sub-agent using a defined prompt template.
[DATA_EXPOSURE]: The skill accesses the user's personal voice profile and corpus located at ~/.claude/memory/USER/VOICE.md and ~/.claude/memory/USER/VOICE_CORPUS/. This is a core feature for style calibration but involves reading sensitive personal writing history.

chronicle