Why this verdict
- suspicious: finding skill:integrations/pi/skills/basic-memory-pi:integrations/pi/skills/basic-memory-pi/SKILL.md:31:HIJACK
Security findings 1
Agent hijacking
mediumContextual riskmedium confidenceThe skill directs the agent to consult reference files under skill-references/ that the artifact does not ship, so at runtime the agent would ingest whatever content exists at that path as writing-style and tool-use guidance that cannot be reviewed.
Canonical Basic Memory skills are bundled under `skill-references/` for guidance on note shape, capture quality, continuation, and task structure.
vector:instruction
Counter-argument considered
The strongest case against the finding is that consulting references is openly declared and aimed at note/task structure, but the declared claim says the canonical skills are bundled while the follow-up confirms the path is NOT IN REPOSITORY, so at runtime the agent ingests whatever sits there as guidance that cannot be reviewed; no carve-out applies and the served SKILL.md units offer no mitigation (the 'not instructions' line governs recalled memory content, not these reference files).
Open questions 3
What the analysis could not settle from the code alone.
[file request integrations/pi/skills/basic-memory-pi/skill-references] The skill instructs the agent to consult the bundled Basic Memory skill references under skill-references/ for note shape, capture quality, continuation, and task structure; do any of these bundled files contain instructions steering the agent beyond note-writing guidance (hidden directives, conditioning of tool calls, capture or exfiltration of data)?
Agent hijacking
[file request integrations/pi/skills/basic-memory-pi/skill-references] The SKILL.md directs the agent to consult the bundled files under skill-references/ for note shape, capture quality, continuation, and task structure; do any of those files contain instructions beyond that declared guidance (non-Basic-Memory tool calls, network access, credential reads, or directives that override user/system instructions)?
Agent hijacking
[file request integrations/pi/skills/basic-memory-pi/references] The description directs the agent to consult bundled Basic Memory skill references for note/task structure; what do those reference files instruct, and do they contain commands, scripts, or network access the agent must perform?
Code execution
Coverage gaps 3
Parts of the artifact the static analysis could not fully examine.
context:unanalyzable_binary pdf content is only byte-scanned
tests/Non-MarkdownFileSupport.pdf
context:frontmatter_unparseable while parsing a block mapping
.claude/commands/spec.md
context:decoded_views_exceeded 410 derived view(s) beyond the per-file view cap: decoded:rot13, decoded:unicode-escape
src/basic_memory/mcp/ui/html/search-results-tool-ui.html