The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that attempt to override standard agent caution and user oversight.
Evidence: The 'Auto Mode' section and the flag --yes instruct the agent to 'Skip all confirmations' and 'skip validation'. Additionally, the 'Core Rules' explicitly state 'DO NOT STOP: Execute until all mappers complete or fail', which may cause the agent to ignore safety signals or errors encountered during execution.
[PROMPT_INJECTION]: Vulnerability to indirect prompt injection through analysis of untrusted codebase data.
Ingestion points: Mapper tasks involve scanning file content using Read, Glob, and Grep tools on sensitive files like package.json and CI/CD configurations.
Boundary markers: The tasks.csv schema and mapper instructions lack clear delimiters or warnings to ignore instructions embedded within the analyzed code.
Capability inventory: Sub-agents spawned by the skill have access to powerful tools including Bash, Write, Edit, and spawn_agents_on_csv.
Sanitization: There is no evidence of sanitization or filtering of the codebase data before it is processed by the AI agents.
[COMMAND_EXECUTION]: Use of shell commands for session and directory management.
Evidence: The skill implementation uses the Bash tool to execute mkdir -p for session folders and output directories.

spec-map