The Agent Skills Directory

[PERSISTENCE]: The skill configures a PostToolUse hook in .claude/settings.json. This mechanism ensures that a shell script (post-edit-check.sh) is automatically executed following every Edit or Write tool invocation within the project. This establishes a persistent execution path that remains active across all agent sessions for the project.
[COMMAND_EXECUTION]: The scaffolding process involves executing several shell commands via the agent's subprocess capabilities, including directory creation (mkdir), file duplication (cp), and permission modification (chmod).
[PRIVILEGE_ESCALATION]: The instructions direct the agent to grant executable permissions (chmod +x) to the post-edit-check.sh script. This script is then integrated into the agent's automated workflow via settings, allowing it to run with the same privileges as the agent without further user confirmation.
[DATA_EXFILTRATION]: The skill accesses and writes to sensitive locations outside the project root, specifically the ~/.claude/ global configuration directory. It calculates a project-specific path within this directory to store and retrieve MEMORY.md, which contains session-specific data and lessons that are automatically loaded into the agent's system prompt.
[INDIRECT_PROMPT_INJECTION]: The skill implements a feature where it reads untrusted project files (like README.md or manifest files) and interpolates their content into CLAUDE.md and MEMORY.md. These files are subsequently used as part of the agent's system prompt or high-priority context in future sessions.
Ingestion points: Root directory file listing, README.md, package.json, pyproject.toml, and the scripts or [tool] sections of manifest files.
Boundary markers: None. The interpolated content is placed directly into templates without delimiters or "ignore instructions" warnings.
Capability inventory: Shell command execution (bash, cp, chmod), directory creation, file writing, and environment variable modification.
Sanitization: No sanitization, escaping, or validation of the ingested content is performed before interpolation into system-level instruction files.

Project Scaffolding