The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources including pipeline.log, iter-*.log, and progress.md to perform its analysis and generate recommendations.\n
Ingestion points: Processes various log files and documentation files across multiple phases (Phase 2, 3, 4, 9).\n
Boundary markers: The instructions lack explicit delimiters or warnings to ignore potentially malicious instructions embedded in the log files being parsed.\n
Capability inventory: The skill is granted Bash, Edit, and Write tools, providing a significant capability surface for file system modification and command execution.\n
Sanitization: No sanitization or validation of the ingested log content is performed before it is used to generate report content and suggested patches.\n- [DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file utilizes the !command syntax to execute shell commands at load time. These commands (git branch, git log, git diff) are used to retrieve local repository state to populate the agent's context.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates code diffs and patches based on its analysis of logs and project state, then applies them to the filesystem using the Edit tool. It also identifies and executes build and test scripts discovered in project configuration files like package.json or CLAUDE.md.\n- [COMMAND_EXECUTION]: The skill invokes shell commands via the Bash tool to perform git operations, log parsing with sed and grep, and execution of the project's own build and test scripts.\n- [OBFUSCATION]: The skill uses hex escape sequences (\x1b) within a sed command pattern to identify and remove ANSI color codes from iteration logs. This is a common administrative task for log cleanup and does not appear to hide malicious intent.

solo-retro