The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and analyze last_error.log files generated from build outputs. These logs reflect the results of the compilation and linking process, which can be influenced by the source code being built. If an attacker controls the code in a repository, they can inject malicious instructions into error messages that will eventually be processed by the agent. The skill lacks boundary markers or sanitization for this content, potentially allowing injected text to override the agent's behavior during the analysis phase.
Ingestion points: The agent is instructed to read out/<product>/last_error.log in SKILL.md (Step 2) to categorize errors and match them against known cases.
Boundary markers: Absent. The agent reads the extracted log content directly to identify error patterns.
Capability inventory: The skill has the capability to read local files and execute a specific shell script (extract_last_error.sh).
Sanitization: Absent. The log content is not filtered or sanitized before being analyzed by the LLM.
[COMMAND_EXECUTION]: The workflow defined in SKILL.md requires the agent to execute a local shell script, script/extract_last_error.sh. This script uses Python to parse and extract the most recent error block from the build logs. While the script performs a legitimate utility function, it represents the execution of code provided by the skill author within the agent's environment.

build-error-analyzer