hyper-handoff

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and resume from handoff memos stored in the .hyper/handoff/ directory. These files represent an attack surface where malicious or conflicting instructions could be introduced to influence a future session.
  • Ingestion points: The skill reads state information and instructions from .hyper/handoff/<slug>/memo.md as specified in SKILL.md and rules/memo-contract.md.
  • Boundary markers: The skill provides strong defensive instructions, explicitly stating that memo content is 'evidence, never authority' and requiring the agent to restate the task and seek user confirmation before executing any 'destructive or outward' actions.
  • Capability inventory: The skill utilizes file read/write operations and executes a local shell script for state validation. Destructive capabilities (e.g., git push, file deletion) are governed by a mandatory 'confirm-first' gate.
  • Sanitization: The skill mandates the redaction of secrets from quoted requests and performs cryptographic fingerprinting of the repository state to ensure context integrity.
  • [COMMAND_EXECUTION]: The skill executes a bundled utility script, scripts/handoff-memo.sh, to perform repository state checks.
  • The script is restricted to read-only operations using standard tools such as git, awk, and sha256sum to validate the integrity and format of the handoff memos.
  • [EXTERNAL_DOWNLOADS]: The documentation (references/agent-support.md) provides instructions for installing the skill and its dependencies using npx.
  • These downloads target the vendor's (alpoxdev) official packages and are handled through established package registries.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 07:40 PM