The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It ingests untrusted data from the codebase (code, git history, documentation) across five different assessment phases (CTO, CFO, CPO, Sales, and Health reviews). These inputs are interpolated into prompts to generate a consolidated investor brief. While it uses defined phases, it lacks explicit boundary markers or instructions to ignore embedded malicious directions within the processed files. The capability inventory includes the ability to write reports to the filesystem (docs/investor-brief.md) and execute sub-skills.
[DATA_EXFILTRATION]: The 'SELF-EVOLUTION TELEMETRY' section attempts to write execution metadata, including outcomes and suggestions, to a local directory (~/.claude/projects/skill-telemetry.md). While this is local, logging execution details from potentially sensitive due diligence reviews into a centralized telemetry file represents a minor data exposure risk if that file is later processed or exfiltrated by other automated 'self-evolution' processes.
[COMMAND_EXECUTION]: The skill uses $ARGUMENTS directly within its execution flow. While intended for context (e.g., 'Series A prep'), if improperly handled by the underlying agent platform, this could lead to command injection if the arguments contain shell metacharacters or agent-specific command overrides.
[SAFE]: The skill references several other skills (/cto-review, /cfo-review, etc.). This modular design is a standard practice for complex task orchestration and does not inherently increase risk beyond the risks of the individual sub-skills.

investor-ready