The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a command-line interface tool called codex to execute tasks and handle background processing.
Evidence: Use of commands like codex exec -m gpt-5.6-sol "reply OK" and backgrounding patterns codex exec ... </dev/null &.
[EXTERNAL_DOWNLOADS]: The skill is designed to send local context and code diffs to external providers (specifically OpenAI) for processing.
Evidence: Instructions in REFERENCE.md describe authorization gates for sending code to OpenAI and cross-provider review mechanisms.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository data and objectives to generate prompts for external execution, which presents a surface for instructions embedded in local files to influence the delegated agent.
Ingestion points: Repository content, branch names, objectives, and scope definitions.
Boundary markers: The skill uses a "Prompt contract" to define scope and exclusions, and instructs the agent to "exclude secrets."
Capability inventory: The skill has the ability to execute shell commands via codex exec and write output to local paths.
Sanitization: It requires verifying citations, commands, and conclusions before integration.

codex