The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill automatically ingests data from the project environment, including git logs, project documentation (README, playbooks, ICP docs), and external sources such as CRM exports and meeting notes to populate a context artifact. This creates a surface for indirect prompt injection where malicious instructions embedded in these data sources could influence agent behavior.
Ingestion points: SKILL.md (Section 1) instructs the agent to read git logs, README files, playbooks, agent-instruction files, and CRM/sequencer exports.
Boundary markers: No explicit delimiters or instructions to ignore embedded content within ingested files are provided.
Capability inventory: The skill possesses the capability to write to sales-context.md and modify existing project agent-instruction files (Section 5).
Sanitization: No evidence of sanitization, validation, or filtering of the ingested external content was found.
[DATA_EXFILTRATION]: The skill performs an automated inventory of the environment that accesses potentially sensitive files, including the contents of the .github/ directory and other agent-instruction files (the system prompts for other skills).
Evidence: SKILL.md (Section 1) explicitly directs the agent to inventory "agent-instruction files" and the ".github/" directory.
[COMMAND_EXECUTION]: The skill utilizes shell commands to access the repository's git history to analyze project activity.
Evidence: SKILL.md (Section 1) instructs the agent to "read the recent log" from the repository's git history.
[PERSISTENCE]: The skill maintains project state across different sessions by creating and updating a sales-context.md artifact and by modifying the agent's own instruction files to persist project invariants.
Evidence: SKILL.md (Section 5) details the creation of the context artifact and the optional patching of the agent-instruction file to store state.

sales-kickoff