The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's 'On Activation' sequence executes arbitrary commands defined in the workflow.activation_steps_prepend and workflow.activation_steps_append arrays within the config.yaml file. This allows for runtime execution of logic defined in persistent configuration.
[DYNAMIC_EXECUTION]: The on_complete hook in customize.toml allows the execution of arbitrary commands after the reporting stage is finished. The command is invoked with a result path argument and executed in a subprocess.
[DYNAMIC_EXECUTION]: The helper script scripts/skf-new-file-diff.py uses importlib.util to dynamically load and execute an external Python module (skf-classify-changed-files.py) from a calculated path.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from source code and authoritative documentation files (e.g., llms.txt, AGENTS.md) to update skill instructions. While it uses AST-based structural analysis to minimize risk, it remains susceptible to malicious instructions embedded in the processed files that might influence the agent's behavior during the update workflow.
Ingestion points: references/re-extract.md (source code parsing) and references/detect-changes.md (authoritative file discovery).
Boundary markers: The instructions lack explicit, robust delimiters or 'ignore' warnings for the content being extracted into the prompt.
Capability inventory: The skill can write to the filesystem using Write/Edit tools and execute local scripts via uv run.
Sanitization: Relies on specific AST recipes to filter and structure content, providing a degree of separation between raw input and agent instructions.

skf-update-skill