The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the agent to operate in 'AUTONOMOUS MODE' and 'Do NOT ask questions,' which suppresses standard user confirmation prompts and oversight for the agent's actions.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted source code as its primary input without sufficient security boundaries.
Ingestion points: The skill reads files and directories provided via the $ARGUMENTS variable in the SKILL.md file.
Boundary markers: Absent; the target code is processed without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can execute shell commands (bash, sed), development tools (flutter, dart, tsc, docker-compose), and perform file system writes and git commits.
Sanitization: Absent; there is no validation or escaping of the content found within the analyzed code files.
[COMMAND_EXECUTION]: The skill executes multiple CLI tools including bash, sed, flutter, tsc, and docker-compose based on the project structure it encounters, providing an execution path for instructions discovered during the review process.
[DATA_EXFILTRATION]: The skill accesses the user's home directory at ~/.claude/projects/ to append telemetry and outcome data to a skill-telemetry.md file, which involves reading and writing to platform-specific configuration paths.

iterate-review