AGENT-INJECTION PRE-SCAN REPORT
================================
files scanned : 4
findings      : 11

[HIGH  ] PI-OVERRIDE        SKILL.md:9
         instruction-suppression phrase — attempts to replace or void the operator's instructions (inside HTML comment — hidden from rendered view)
[HIGH  ] PI-SECRET-ACCESS   SKILL.md:10
         credential-access imperative — tells the agent to read key/secret material (inside HTML comment — hidden from rendered view)
[HIGH  ] PI-CONCEAL         SKILL.md:11
         concealment directive — instructs the agent to hide actions from the human (inside HTML comment — hidden from rendered view)
[HIGH  ] PI-OVERRIDE        mcp.json:3
         instruction-suppression phrase — attempts to replace or void the operator's instructions
[HIGH  ] PI-CONCEAL         mcp.json:3
         concealment directive — instructs the agent to hide actions from the human
[MEDIUM] PI-AUTHORITY       SKILL.md:9
         forged-authorization phrase — claims approval/consent the file cannot prove (inside HTML comment — hidden from rendered view)
[MEDIUM] PI-TOOL-ABUSE      mcp.json:3
         destructive/financial imperative embedded in instruction text
[MEDIUM] PI-HIDDEN-UNICODE  notes.md:3
         invisible text carriers: U+200B ZERO-WIDTH-SPACE x1
[MEDIUM] PI-TOOL-ABUSE      notes.md:4
         destructive/financial imperative embedded in instruction text
[MEDIUM] PI-B64-BLOB        notes.md:8
         base64/base64url blob decoding to readable prose — common channel for smuggled instructions past human review
[LOW   ] PI-EVAL            notes.md:10
         decode-and-execute imperative — obfuscated-payload delivery

Scope: deterministic pre-scan only — flags are for human review, not a verdict.