The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The run-shipped-checkpoints.sh script clones the automated-assessment-skill repository and other skill repositories from GitHub to perform maturity audits. It also downloads the yq utility from its official GitHub releases.
[REMOTE_CODE_EXECUTION]: The skill executes the run-checkpoints.sh script fetched from the remote automated-assessment-skill repository to evaluate project checkpoints.
[COMMAND_EXECUTION]: The skill extensively uses shell commands and utilities including git, grep, awk, sed, yq, and jq to verify repository consistency and documentation accuracy.
[PRIVILEGE_ESCALATION]: Provided CI workflow templates (harness-checkpoints.yml.tmpl, forgejo-harness-checkpoints.yml.tmpl) contain instructions to use sudo for installing system dependencies like yq and jq.
[INDIRECT_PROMPT_INJECTION]: The skill implements an attack surface for indirect prompt injection by processing .harness/checkpoints.yml, which defines remote repositories to be cloned and audited.
Ingestion points: .harness/checkpoints.yml (skill repository and ref declarations).
Boundary markers: Standard YAML formatting and pinned version references.
Capability inventory: git clone operations, shell script execution via the assessment runner.
Sanitization: Input variables are quoted in shell commands to prevent basic word splitting and injection.
[DYNAMIC_EXECUTION]: The verify-harness.sh script uses inline Python commands to parse composer.json and package.json files for script validation. Additionally, checkpoints.yaml defines complex Bash scripts that are executed by the assessment runner.

agent-harness