The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and follow instructions stored in local markdown files within the docs/knowledge/ directory. If these files are manipulated by an external contributor (e.g., via a Pull Request), they could contain malicious instructions designed to bypass constraints or influence the agent's logic.
Ingestion points: Reads docs/knowledge/INDEX.md and specific notes in docs/knowledge/.
Boundary markers: Absent. The agent is instructed to treat these notes as authoritative "prior lessons" and "hard constraints".
Capability inventory: Performs file reads, shell commands via grep, and influences future decision-making based on note content.
Sanitization: No sanitization or validation of the note content is performed.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to use shell commands to search through documentation.
Evidence: grep -ril "<error text>" docs/knowledge/

dev-knowledge