The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run to execute external commands. It uses the GitHub CLI (gh) to fetch metadata, job details, logs, and issue information. It also uses curl to fetch HTML from GitHub Actions to identify workflow syntax errors. All calls use argument lists rather than shell strings, and user-provided inputs like run IDs are validated as digits before being passed to commands.
[EXTERNAL_DOWNLOADS]: The skill downloads content from external sources to perform its diagnostics. It retrieves job logs via the GitHub API and scrapes HTML content from github.com. These operations are limited to the official GitHub domains and the vendor's specific repository (dotCMS/core).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external GitHub logs and workflow annotations, which represents a surface for indirect prompt injection.
Ingestion points: Raw logs are downloaded via github_api.py and workflow annotations are scraped from HTML via html_scraper.py.
Boundary markers: Evidence is presented to the agent with clear section headers and dividers (e.g., === FAILURE EVIDENCE FOR ANALYSIS ===) in evidence.py and diagnose.py.
Capability inventory: The skill can execute shell commands via gh and curl and write files to a local workspace (.claude/diagnostics).
Sanitization: The skill performs basic HTML tag removal using regex in diagnose.py and uses tiered extraction to filter and truncate logs to specific relevant sections, which limits the influence of external content.

dot-cicd-diagnose