The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a workflow for diagnosing build failures where it parses the build.log file, extracts the command following a FAILED: marker, and instructs the agent to execute it directly in the shell. This creates a dependency on external file content for command construction during the debugging phase.
[PROMPT_INJECTION]: The skill's log parsing logic is susceptible to indirect prompt injection.
- Ingestion points: The src/out/<product>/build.log file is processed by scripts/analyze_build_error.sh and SKILL.md instructions.
- Boundary markers: No explicit delimiters or warnings are used to prevent the agent from obeying instructions embedded in the log file.
- Capability inventory: The agent has the capability to execute shell commands, write files (snapshots), and modify source code within the project directory.
- Sanitization: There is no validation or escaping performed on the strings extracted from the log before they are suggested for execution.

arkweb-build