The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions by reading and applying insights from previously recorded observations.
Ingestion points: Frontmatter and bodies of Markdown files located in [workspace folder]/skill-observations/observation-log/ (e.g., in SKILL.md and references/observation-log.md).
Boundary markers: Relies on standard YAML frontmatter delimiters (---).
Capability inventory: The skill utilizes shell commands (find, grep, awk, sed, mkdir, cp, chmod, printf, mv) and Python scripts to process these logs.
Sanitization: Instructions in references/environments.md direct the agent to "Apply their insights to the current work," which involves processing the natural language content of the logs. While not explicitly sanitized, the data source is the agent's own prior output or directed user feedback, aligning with its core purpose of continuous improvement.
[COMMAND_EXECUTION]: The skill provides numerous shell scripts and snippets for log management and validation.
Evidence: Includes a complex awk program for YAML validation in SKILL.md, an ID derivation and archival logic in references/logging.md, and utility scripts such as scripts/new-observation.sh. These commands are strictly scoped to the local observation workspace and serve established functional requirements for the logging system.

task-observer