The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute standard shell commands including pwd, git status, and git log to identify the current working directory and verify recent project activity. These commands are necessary for the skill's stated purpose and do not pose a security risk in this context.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files including .claude/PROJECT_CONTEXT.md, CLAUDE.md, package.json, and requirements.txt. While this creates a surface for indirect prompt injection if these files contain malicious instructions, it is the intended primary purpose of the skill to restore context from these sources.
Ingestion points: Files like PROJECT_CONTEXT.md, CLAUDE.md, and technical manifests are read into the agent's context (e.g., in SKILL.md and docs/2025-11-30-smart-context-loader-design.md).
Boundary markers: The skill does not explicitly define delimiters to separate this external content from its own instructions.
Capability inventory: The skill has the capability to read/write files and execute git commands based on the context loaded.
Sanitization: No specific content filtering or sanitization is mentioned for the ingested file data.
[PERSISTENCE]: The skill implements a persistence mechanism to maintain state between Claude sessions. It writes updated task lists and status information to .claude/PROJECT_CONTEXT.md and logs metadata to ~/.claude/skill-analytics/last-outcome-project-context-skill.json. This behavior is transparent and aligned with the tool's goal of progress tracking.

project-context-skill