The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill employs eval and source commands to execute the output of scripts located in ~/.claude/skills/gstack/bin/ (such as gstack-slug and gstack-repo-mode). This allows for dynamic modification of the execution environment at runtime based on the output of local scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from plan files and design documents into a multi-phase execution pipeline.
Ingestion points: The skill reads plans and design docs from ~/.gstack/projects/ and the local worktree.
Boundary markers: Includes an instruction for the codex subagent to ignore SKILL.md files, but lacks comprehensive delimiters for all processed data across the CEO, Design, and Engineering phases.
Capability inventory: The skill has access to Bash for command execution and Write/Edit for filesystem modifications across the repository.
Sanitization: No explicit sanitization or filtering of the ingested plan content is performed before processing by subagents.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to run local utility scripts, manage project metadata, and interact with external CLI tools like codex, git, gh, and glab.
[DATA_EXFILTRATION]: The skill collects and logs environment metadata, such as repository and branch names, along with usage telemetry. This information is gathered for local logging and can be transmitted to external endpoints through gstack-telemetry-log or synced to remote GitHub repositories via gstack-brain-sync.
[EXTERNAL_DOWNLOADS]: The skill includes logic to check for software updates and sync artifacts across different machines, requiring network access and the retrieval of content from external servers.

autoplan