The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill executes output from local vendor binaries using eval and source commands (e.g., gstack-slug, gstack-repo-mode in ~/.claude/skills/gstack/bin/). This enables dynamic configuration and behavior based on the local environment.
[REMOTE_CODE_EXECUTION]: The skill uses OpenAI's Codex CLI (codex exec) to perform independent documentation audits. This involves executing review logic on a remote service using content from the local repository.
[EXTERNAL_DOWNLOADS]: The skill references and suggests the installation of the @openai/codex package from the official NPM registry to support its review functionality.
[COMMAND_EXECUTION]: The skill relies on multiple CLI tools and binaries located in the vendor's application directory (~/.claude/skills/gstack/bin/) to manage telemetry, configuration, and repository analysis.
[DATA_EXFILTRATION]: The skill transmits telemetry data (e.g., usage statistics, durations, and outcomes) to an external server. It also sends repository documentation and diffs to OpenAI's Codex service for analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Documentation files, git diffs, and pull request metadata read from the current project.
Boundary markers: No explicit delimiters are used when passing file contents into the agent's context or to the Codex service.
Capability inventory: The skill has access to shell execution, file system modifications, and network operations through git and cloud platform CLIs.
Sanitization: The skill uses a local redaction tool (gstack-redact) to identify and filter sensitive information before updating external pull request descriptions.

document-release