The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents workflows that ingest data from external sources, specifically GitHub issues, which creates a surface for indirect prompt injection.\n
Ingestion points: examples/ship-issue.local.ts uses gh issue list to fetch issue data that is then passed to an agent step as context via {{steps.find-issue.output}}.\n
Capability Inventory: The agents defined in the examples have the capability to modify the local file system and interact with remote repositories via git and gh.\n
Sanitization: The skill includes a utility in examples/assert-repo.ts to validate repository and branch names used in shell commands, but does not define specific sanitization for the external issue content itself.\n- [DYNAMIC_EXECUTION]: The skill facilitates the creation of 'deterministic' steps that execute shell commands constructed at runtime from variables.\n
Evidence: examples/ship-issue.local.ts demonstrates constructing shell commands for git and gh using repository and branch names.\n
Mitigation: The skill provides an assert-repo.ts module to validate these inputs against shell metacharacters, mitigating command injection risks for these fields.\n- [COMMAND_EXECUTION]: The skill's core functionality relies on the execution of local CLI tools, including git, gh, and the vendor's own relayflows CLI.\n- [EXTERNAL_DOWNLOADS]: The documentation and examples reference the installation of the @relayflows and @agent-relay SDK packages.\n
Evidence: examples/README.md and SKILL.md provide instructions for installing @relayflows/core, @relayflows/cli, and @agent-relay/sdk.\n
Context: These packages are components of the author's own infrastructure and are necessary for the skill's intended functionality.

writing-agent-relay-workflows