The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the load-time execution syntax (!) to resolve a file path. The command echo "${CLAUDE_SKILL_DIR}/../../workflows/mine-claude-md-from-sessions.workflow.js" uses a platform-standard environment variable to identify the location of a script bundled with the skill. This is a benign use of dynamic context for configuration.
[DYNAMIC_EXECUTION]: The skill uses the Workflow tool to execute a JavaScript file whose path is determined at runtime. This script contains the primary processing logic for the skill. The target is a vendor-provided resource distributed as part of the skill package, which is consistent with the skill's documented orchestration purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from historical session logs and project files.
Ingestion points: The workflow reads data from the project's session store and repository files within SKILL.md.
Boundary markers: While specific delimiters are not defined in the instructions, the workflow establishes a logical boundary by separating the mining phase from the verification phase.
Capability inventory: The skill has access to the Workflow tool and Bash shell.
Sanitization: The workflow includes a dedicated verification phase where an independent "skeptic" component adversarially tests each mined candidate for validity and novelty before the information is presented to the user.

mine-claude-md