/looptrip

Run the looptrip detector over the given source and report any multi-agent coordination pathologies it finds. looptrip is a read-only observer — it never modifies state, never blocks, and never proposes that you kill or gate anything. It only reports what it sees.

Source to scan: $ARGUMENTS

A source must be one of looptrip's source specs (it is not a bare file path):

• cast-db:<session_id> — handoffs from the local CAST observability DB (~/.claude/cast.db)
• otel:<path>[#scenario] — an OpenTelemetry GenAI handoff span export (JSON or JSONL)
• fixture:<session_id> — a packaged hermetic fixture; the session id is a full UUID, e.g. fixture:da27b414-f9f1-4c91-bd50-1a6096555066 (one of the two $792.96 proof-run sessions)

Steps:

1. Run looptrip scan --all $ARGUMENTS via Bash. (--all runs all four detectors: duplicate_work, ping_pong, deadlock, non_termination.) The looptrip CLI must be installed — pip install looptrip (published on PyPI). A Homebrew formula is also available once the public tap is published: brew install ek33450505/looptrip/looptrip.
2. If the source spec is missing or malformed, show the user the valid forms above and stop.
3. Summarize the findings table: for each pathology that tripped, give the kind, the signature, and where it first tripped. If nothing tripped, say so plainly.
4. Do not take any action on the findings — no fixes, no kills, no config changes. looptrip observes; deciding what to do about a loop is the human's call.