The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user-provided decisions and writes them to persistent documentation files (docs/adr/ and CONTEXT.md).
[INDIRECT_PROMPT_INJECTION]: The instruction 'Take the decision as stated — no interview, no grilling' explicitly directs the agent to bypass critical evaluation or verification of the provided content. If the recorded decision contains malicious instructions, they could influence future agent actions when the project documentation is later retrieved into the agent's context.
[INDIRECT_PROMPT_INJECTION]: Evidence Chain for Ingestion Surface:
Ingestion points: User decision content provided in response to SKILL.md triggers.
Boundary markers: Absent; input is accepted 'as stated' without delimiters.
Capability inventory: File system write access to docs/adr/ and CONTEXT.md.
Sanitization: No validation or sanitization is performed on the user-supplied decision string.

record-decision