The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides automated setup for the bun runtime. It downloads the installer from its official domain (bun.sh) and performs a verification of the script's SHA-256 checksum against a hardcoded value before execution to ensure the integrity of the downloaded content.
[REMOTE_CODE_EXECUTION]: The skill can optionally send project context to the codex CLI for an external AI-driven design review. This capability is protected by a user confirmation step (AskUserQuestion), preventing unauthorized data transmission to external services.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the local filesystem, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Local files including CLAUDE.md, TODOS.md, version control history via git log, and existing design documentation in ~/.gstack/projects/.
Boundary markers: No specific delimiters or delimiters were observed to isolate ingested content or instruct the model to ignore embedded commands.
Capability inventory: The agent has permissions for filesystem modification (Write, Edit), shell access (Bash), and external network interactions (WebSearch, codex).
Sanitization: No evidence of sanitization or filtering of read file content was found before the data is used in prompts.
[DYNAMIC_EXECUTION]: The preamble and various phases of the skill use eval and source on the output of local binaries (e.g., gstack-slug, gstack-repo-mode). This allows the skill to dynamically adjust its configuration based on the specific project and environment it is running in.

office-hours