The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill maintains educational state by reading from and writing to a local file, which introduces a surface for indirect prompt injection.
Ingestion points: As described in SKILL.md (Phase 1) and the Read rules in PROGRESS.md, the agent is instructed to load user state from the file ~/.socratic-tutor/progress.md at the start of every session.
Boundary markers: Absent. The skill instructions do not define delimiters or provide "ignore embedded instructions" warnings for the agent when processing the content of the progress file.
Capability inventory: Across SKILL.md and PROGRESS.md, the agent is granted capabilities to read and write files using available tools, specifically to manage the progress file and create its parent directory.
Sanitization: Absent. There is no evidence of validation or content filtering to ensure that the data read from the progress file does not contain malicious instructions that could influence the agent's behavior.
[TIME_DELAYED_CONDITIONAL]: The skill uses date-based logic to trigger specific teaching modes.
Evidence: SKILL.md (Phase 1.5) and PROGRESS.md define a review schedule based on the Ebbinghaus forgetting curve, where the agent compares the current date to stored values to determine if a "Review Mode" should be initiated.
[NO_CODE]: The skill is entirely defined within markdown instructions and does not include any executable scripts, binaries, or configuration files for programming languages like Python or Node.js.

socratic-tutor