The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The babysit-build sub-agent and the associated "Monitoring Builds Until Green" workflow are designed to autonomously diagnose and fix failures by analyzing external build data. This creates a surface for indirect prompt injection.
Ingestion points: Build logs (teamcity run log --raw) and test results (teamcity run tests --failed) as documented in _agents/babysit-build.md and references/workflows/fix-and-verify.md.
Boundary markers: Absent; the instructions do not specify any delimiters or warnings to ignore instructions embedded within the logs or test output.
Capability inventory: The agent has access to Bash, Read, Edit, Write, Grep, Glob tools, allowing it to modify source code, update DSL configurations, and execute shell commands.
Sanitization: Absent; the agent is explicitly instructed to perform "Code and DSL fixes (repo changes) — act immediately, no confirmation needed."
[REMOTE_CODE_EXECUTION]: The skill documents and enables the use of teamcity agent exec and teamcity agent term, which allow for executing arbitrary commands or opening interactive shells on remote TeamCity build agents. While these are legitimate administrative features of the TeamCity platform, they provide a direct path for the AI to execute code on remote infrastructure.
[COMMAND_EXECUTION]: The skill extensively uses the teamcity CLI to drive operations including build orchestration, agent management, and project configuration changes.
[CREDENTIALS_UNSAFE]: The documentation for teamcity auth login in references/commands.md identifies an --insecure-storage flag. When used, this flag causes the CLI to store sensitive access tokens in plain text configuration files instead of using the secure system keyring.

teamcity-cli