The Agent Skills Directory

[SAFE]: The skill implements standard Harness pipeline configuration logic using authorized MCP tools. No indicators of exfiltration, unauthorized command execution, or malicious intent were found. All external references point to official Harness documentation and the production platform.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes existing pipeline configurations (yamlPipeline) via harness_get, which presents an attack surface for indirect prompt injection. Mitigation: The skill mandates 'Surgical YAML only' updates (Rule 20), strictly limiting the agent to inserting or updating the SscaArtifactSigning step and CI failure strategies while explicitly forbidding reordering or the addition of unrelated security scanners. Evidence: Ingestion points: Pipeline YAML fetched in Phase 1 (SKILL.md); Boundary markers: Instructions to ignore unrelated steps and follow surgical edit rules; Capability inventory: pipeline write access via harness_update; Sanitization: Explicit restrictions against reconstruction of the full pipeline.

sign-artifact