The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to audit untrusted third-party content, such as articles and social media posts. This creates a surface for indirect prompt injection, where malicious instructions embedded in the analyzed text could attempt to override the agent's audit instructions or trick it into providing a specific verdict. The skill mitigates this through explicit warnings and a structured evaluation process, such as the 'four-axis inference boundary' and the directive that 'Authorship not established from prose cues', but the ingestion of untrusted data remains an inherent risk.
Evidence of ingestion points: The scripts/corpus_repetition.py script and the synthesis-content-quality methodology ingest and process external Markdown files.
Evidence of capability inventory: The skill uses Python scripts to perform word-frequency analysis and repetition detection, providing a processing surface for untrusted strings.
[COMMAND_EXECUTION]: The skill includes utility and test scripts that utilize subprocess.run to execute shell commands, which is a powerful capability that requires caution.
Evidence in tests/generate_no_removals_baseline.py: Employs subprocess.run to execute git commands (git ls-tree, git show) on a specified repository path.
Evidence in tests/test_corpus_repetition.py: Employs subprocess.run to execute the corpus_repetition.py script within its test suite.

synthesis-content-quality