The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes eval and source with process substitution to execute the output of local binaries located in ~/.claude/skills/gstack/bin/ (e.g., gstack-slug, gstack-repo-mode). This mechanism is used to dynamically modify the agent's shell environment variables and state at runtime.
[COMMAND_EXECUTION]: The skill executes an extensive shell preamble to perform environment discovery, session management, and update checks. It also possesses the capability to modify the project's CLAUDE.md file and perform git commits if the user opts into automated skill routing rules.
[DATA_EXFILTRATION]: The skill implements a telemetry and analytics framework that logs usage data (such as skill name, session duration, and repository name) to the local filesystem. If the user opts in, this data is transmitted to an external service via the gstack-telemetry-log binary.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and displays the contents of previously saved context files and learning logs from the project directory. These files are potentially untrusted as they may contain data from external sources or past sessions that could influence the agent's behavior.
Ingestion points: Reads from ~/.gstack/projects/${SLUG}/checkpoints/*.md, learnings.jsonl, and timeline.jsonl.
Boundary markers: The restored context is presented with visual delimiters (e.g., ════) and headers, but does not include explicit instructions for the agent to ignore directives embedded within the restored text.
Capability inventory: The skill has access to shell execution (Bash) and file system access tools (Read, Glob, Grep).
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from the context checkpoints before it is presented to the agent.

context-restore