The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from CI environments, including PR check results, workflow outcomes, and log files, which may contain adversarial content designed to influence the agent's behavior.
Ingestion points: The skill reads PR-scoped data, CI failure logs, and entries from .learnings/HEALS.md (SKILL.md).
Boundary markers: No specific boundary markers or sanitization logic are defined to delimit untrusted data within the prompts.
Capability inventory: The skill emits structured learning output and proposes repository modifications (such as updates to CLAUDE.md or AGENTS.md) via pull request comments or new PRs using the GitHub CLI (SKILL.md).
Sanitization: No explicit sanitization or validation of the distilled rules or captured patterns is mentioned before they are proposed as promotion candidates.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install external tools and extensions.
GitHub Extension: Fetches the gh-aw extension from GitHub's official organization (SKILL.md).
Skill Installation: Downloads the skill and its dependencies using gh skill install and npx skills add from the vendor's repository (SKILL.md).
[COMMAND_EXECUTION]: The skill relies on shell commands for installation, validation, and execution within CI pipelines.
Tool usage: Utilizes gh, npx, and gh-aw to manage skills, compile workflows, and trigger test runs (SKILL.md, references/workflow-example.md).

self-improvement-ci