The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on multiple Bash snippets and Python scripts (migrate-log.py, new-observation.sh, validate-skill-bundle.py) to automate log management, ID generation, and skill validation. These scripts are invoked by the agent to maintain the workspace and provide structural enforcement for its data-logging rules.
[EXTERNAL_DOWNLOADS]: The skill-authoring.md reference specifies a protocol for 'feedback pre-flight' that uses curl to fetch the 'upstream HEAD' of files from GitHub repositories (e.g., github.com/rebelytics/*) for comparison. This is intended to ensure local changes are verified against the latest official source before submitting contributions.
[PERSISTENCE]: The weekly-review.md procedure includes a process for setting up scheduled review tasks using system-level schedulers such as cron on Unix, launchd on macOS, or Windows Task Scheduler. This mechanism allows the skill's review logic to persist and run periodically as a background task.
[INDIRECT_PROMPT_INJECTION]: As a meta-skill that analyzes other skills and project metadata (like CLAUDE.md), it possesses a significant data ingestion surface for external/untrusted content. However, the skill provides clear guidance on using boundary markers (YAML frontmatter) and structural verification steps to mitigate the risk of accidental instruction obedience.

task-observer