The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it requires the agent to read and triage external Pull Requests, issues, and discussions. An attacker could embed malicious instructions in these inputs to influence the agent's behavior during sensitive operations like merging or release preparation.- Ingestion points: The skill instructions in SKILL.md (sections 'Source Checks' and 'Maintainer Sweep') direct the agent to inspect live PRs, issues, and discussions.- Boundary markers: While the skill enforces structural boundaries (such as requiring a 'clean maintainer checkout' and exact SHA attestations), it lacks explicit textual delimiters to prevent the agent from following instructions embedded within the PR data.- Capability inventory: The skill has the authority to execute repository-altering commands, including npm run merge:batch, npm run release:prepare, npm run release:publish, and git push.- Sanitization: The skill relies on external validation scripts (npm run validate, npm run security:docs) to detect policy violations, but these do not sanitize the natural language context against prompt injection.- [COMMAND_EXECUTION]: The skill operates by executing various local npm scripts and shell commands (npm run merge:batch, npm run release:publish, npm run test:aas-v1, git fetch origin/main). These commands provide the agent with significant control over the project's source code and deployment pipeline.

antigravity-maintainer-batch-release