The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a pipeline in 'SKILL.md' that ingests untrusted data via the 'context' parameter provided by the user. There are no explicit boundary markers (e.g., delimiters) to separate this data from instructions, and no sanitization or filtering of the external content is prescribed. While the skill is granted high-capability tools ('Bash', 'Write', 'Edit', 'Read', 'Grep', 'Glob' in 'allowed-tools'), the risk of exploitation is mitigated by the rigid functional constraints that limit output to short, paste-ready UI strings.- [DATA_EXFILTRATION]: No evidence of credential harvesting, access to sensitive file paths like '.env' or '.ssh', or unauthorized network operations was found across the skill's instructions or reference files.- [REMOTE_CODE_EXECUTION]: The skill does not perform remote script downloads, install unverified packages, or execute code from external sources. It relies on internal logic and a peer 'writer' skill.- [OBFUSCATION]: No Base64-encoded strings, zero-width characters, homoglyph attacks, or hidden text patterns were detected in the skill files or metadata.

microcopy