The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external sources.
Ingestion points: Data enters the agent context through GraphQL queries fetching Pull Request titles, labels, and bodies in fetch.md.
Boundary markers: The skill includes an explicit instruction in SKILL.md to treat external content as data rather than instructions, acknowledging the risk of prompt injection attempts in PR titles.
Capability inventory: The skill utilizes gh api for data retrieval, gh gist create for publishing reports, and python3 for setup checks across SKILL.md, fetch.md, and export.md.
Sanitization: While the skill instructs the agent to flag injection attempts to the user, there is no evidence of programmatic sanitization or strict schema validation for the ingested strings.
[Automated Network Communication]: As defined in export.md, the skill is designed to automatically publish generated HTML dashboards to a secret GitHub Gist. This represents an automated outbound network operation that transmits data processed by the agent to a remote service, which is a standard part of this skill's dashboarding functionality.
[Local Command Execution]: The skill executes shell commands, including the GitHub CLI (gh) and python3 modules, to perform GitHub API operations and project pre-flight checks. These executions are necessary for the skill's logic but involve standard command execution patterns.
[Configuration and State Access]: The skill reads and updates local state and configuration files, such as .apache-magpie.lock and .apache-magpie.session-state.json. These actions are used for drift detection and to maintain a stable identifier for the published dashboard across sessions.

stats