Traser: a local deterministic tool to triage agent runs that succeed but return wrong results
Sensitive-Parsnip-12 · reddit · 2026-09-10
A developer released Traser, a local tool for a common agent-debugging pain: the run completes, the trace exists, nothing obviously crashed, yet the outcome is wrong — and someone still spends an afternoon manually digging through execution.
Recurring failure patterns the tool targets:
- previous step looks fine but the next step received different state
- tool call reports success while the response body hides an error or rate limit
- wrong account/id carried forward while every step stays schema-valid
- prompt/model input changes invisible in the trace
- retries eventually succeed via a different execution path
- the configured model/provider isn't the one that actually served the request
- a skipped verification step with everything downstream still completing
Traser takes a suspicious run plus an optional trusted run, normalizes and aligns steps, compares inputs, prompts, tool args, outputs, state, retries and provenance, suppresses low-value diffs, and surfaces a small set of evidence-backed spots worth checking. The engine is deterministic (no LLM), analysis runs locally in the browser, and it deliberately claims no root cause — only investigation leads.
More from coding & agent
- GPT-6 Astra drives Houdini for hands-off procedural modeling with one prompt — ssh4net · 2026-09-10
- SAEScientist-Bench: can AI agents autonomously run SAE interpretability research? — CASIA · 2026-09-10
- Dev uses GPT Astra to port a 33-year-old game to HTML5 with online mode — Vintendopower · 2026-09-10
- mcp-oracle-h adds a mandatory human approval gate for irreversible agent actions via Telegram — modelcontextprotocol · 2026-09-10
- Codex already lets you group pinned sessions by dragging them into projects — vwxyzjn · 2026-09-10
- Use /goal to steer the astra coding agent back on track — TheMoonMidas · 2026-09-10