The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It reads research reports from '~/CybosVault/private/deals//research/*.md' and interpolates the raw content directly into LLM prompts. Ingestion points: Workflow file 'workflows/generate.md' reads all markdown files in the company research directory. Boundary markers: Absent. The prompts for the 'memo-analyst' and 'memo-writer' agents lack delimiters or negative constraints to prevent the model from following instructions embedded in the research reports. Capability inventory: The skill has permissions to read from and write to the filesystem, including overwriting files in the 'memo' directory. Sanitization: No sanitization or validation of the input research content is performed.
[DATA_EXFILTRATION]: The skill accesses and processes data within '~/CybosVault/private/', which contains sensitive investment and deal information. Although no network exfiltration patterns were detected, the handling of private files constitutes a data exposure surface. This access is integral to the skill's stated purpose.
[NO_CODE]: The skill does not include any executable scripts or binary files, relying entirely on markdown instructions and agent workflows.

ddmemo