The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation describes recording shell commands used during development to provide context for 'clues'. The skill itself does not execute these commands; it only logs them as text in Markdown files within the project's '.xgent-ai' directory for later review.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it processes untrusted data from the execution environment and user corrections.
Ingestion points: Reads clues, failure messages, and user corrections from the current session context (SKILL.md).
Boundary markers: The skill enforces a strict 3-4 line Markdown template for entries and initializes them with a 'pending' status.
Capability inventory: File-write access to the local '.xgent-ai/experience/' directory (SKILL.md).
Sanitization: Explicitly prohibits the storage of secrets, tokens, keys, or internal thinking processes. Crucially, it mandates that recorded entries must not be injected into subsequent tasks as rules until a separate 'reflect' process occurs.
[DATA_EXPOSURE]: The skill records file paths and command history as context for captured experiences. It mitigates data exposure risks by instructing the agent to only record 'minimal necessary summaries' and safe location information, excluding any sensitive raw data.

experience-capture