The Agent Skills Directory

[COMMAND_EXECUTION]: The maintenance script scripts/check_v04.py uses subprocess.run to execute other validation scripts within the repository (e.g., scripts/check_sources.py). This is used for automated testing of encoding regressions and verifying script syntax during family-level maintenance.
[DYNAMIC_EXECUTION]: The validation tool scripts/check_v04.py uses runpy.run_path to dynamically load and execute functions from local repository scripts (such as check_prose_quality.py and convert.py). This allows the maintenance workflow to run regression tests on the internal logic of specialized skills.
[OBFUSCATION]: Privacy and source auditing scripts (scripts/check_path_privacy.py, scripts/check_sources.py) use string concatenation for path keywords (e.g., "Us" + "ers", "ho" + "me"). This is a common and benign defensive technique used in security scanners to prevent the scanning tool from accidentally flagging its own source code as a leak.
[DATA_EXPOSURE_AND_EXFILTRATION]: The scripts/check_sources.py script performs HTTP HEAD requests via urllib.request.urlopen to validate the reachability of URLs listed in references/source-registry.yaml. This functionality is restricted to the static list of academic and technology domains defined by the skill and is intended for link maintenance.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data such as manuscripts and PDF reviews. It explicitly mitigates indirect prompt injection risks in references/privacy-and-evidence.md by instructing the agent to ignore any instructions embedded inside processed papers, reviews, or web pages.

ccf-common