The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection by processing historical session data to verify the effectiveness of rules.
Ingestion points: The skill reads data from .reflect/applied.md, recent session transcripts, and the current conversation.
Boundary markers: Absent. The instructions do not require the agent to use delimiters or ignore instructions that may be embedded within the session evidence.
Capability inventory: The agent is authorized to read project files and rewrite configuration files, such as removing lines from AGENTS.md when rules are marked as pruned.
Sanitization: Absent. There is no requirement for validation or escaping of the content found in session transcripts before it is used to grade and prune rules.
[NO_CODE]: The skill consists entirely of natural language instructions in markdown and does not include any scripts, executables, or external package dependencies.

reflect-feedback