The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the persistence of session data into a long-term memory store, creating a surface for indirect prompt injection if the session content contains untrusted or adversarial instructions.
Ingestion points: SKILL.md Phase 1 (Session Analysis) ingests accomplishments, decisions, problems, and file changes from the active session context.
Boundary markers: Absent. The skill does not implement delimiters or 'ignore' instructions when formatting session data into memory templates.
Capability inventory: Employs write_memory and list_memories tools to store and retrieve persistent state.
Sanitization: Absent. There is no evidence of filtering, escaping, or validation of the session content before it is persisted as a learning or session checkpoint.
Impact: Persisted memories, particularly those with the CRITICAL_* prefix defined in references/memory-conventions.md, are intended to be loaded in future sessions, allowing potentially injected instructions to bypass the immediate session context.

save