The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external organizations and release sources to generate instructions (playbooks) for other agents. This creates a surface for indirect prompt injection where malicious content in release notes or organization metadata could influence the generated playbook instructions.
Ingestion points: Metadata and release data fetched via releases admin org get {slug} and releases list {source-slug} in the 'Verified prompt template' (SKILL.md).
Boundary markers: Absent. The prompt templates interpolate data directly without specific delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can spawn sub-agents via the Agent tool and execute local shell commands via the releases CLI and Bun.
Sanitization: No evidence of sanitization or validation of the ingested release data before it is processed by the sub-agent.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic execution patterns to process data and manage sub-agents.
Uses bun -e to execute inline JavaScript strings for formatting and filtering organization data (e.g., Step 1 and Step 6 in SKILL.md).
Dynamically spawns sub-agents using the Agent tool with programmatically generated prompt templates.
[COMMAND_EXECUTION]: The skill relies on local shell command execution for its core functionality.
Extensive use of shell pipes, heredocs (<<'NOTES'), and environment variables to interact with the releases CLI tool.
Modifies the local filesystem by creating and writing to work logs in ~/.releases/work/.

seeding-playbooks