The Agent Skills Directory

[PROMPT_INJECTION]: No instructions were found that attempt to override agent behavior or bypass safety filters.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file access (e.g., .ssh, credentials) or network-based exfiltration patterns were detected.
[OBFUSCATION]: The skill does not use encoding or hidden characters to obscure its instructions or commands.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: No external script execution or third-party package installations are performed.
[PRIVILEGE_ESCALATION]: The instructions do not involve gaining elevated privileges or using sudo.
[PERSISTENCE_MECHANISMS]: No attempts to create persistent access via startup files or scheduled tasks were found.
[METADATA_POISONING]: Metadata fields are accurate and consistent with the functional behavior of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect injection as it ingests local source code for analysis.
Ingestion points: Scans project source files via Read and Grep as defined in SKILL.md.
Boundary markers: Absent.
Capability inventory: Uses Bash, Edit, and Write tools for metadata and handoff management.
Sanitization: No sanitization is performed on analyzed code content before reporting.
[TIME_DELAYED_OR_CONDITIONAL_ATTACKS]: No malicious logic is gated by time or external triggers.
[DYNAMIC_EXECUTION]: The skill does not generate or execute code dynamically.
[DYNAMIC_CONTEXT_INJECTION]: The exclamation-backtick syntax for dynamic context injection is not used in the skill instructions.

time-bomb-radar