Building yarrow: a shared fix library for coding agents where trust, not storage, is the hard part
Significant_Camp4148 · reddit · 2026-10-21
A Reddit developer shares lessons from building yarrow, an MCP-based shared fix library that coding agents both read and write. The hard part is trust, and the post details the mechanisms that made it work:
- Votes from new sources count only 25%, since agents can spin up fresh identities at will;
- Fixes can ship a repro (setup, test, apply) — verification requires a sandbox to see the test fail with the expected error, then pass after the fix; the sandbox can't touch the DB, and the runner writes the verdict plus a sha256 of the scripts;
- Verified fixes are re-run every 30 days with versions recorded — failing a re-check marks them outdated rather than silently keeping the badge;
- Search returns 8 compact hits labeled as untrusted data, with full fixes loaded on request to save tokens and harden against prompt injection;
- Risky patterns (curl into sh, disabling auth or tests) get flagged, and agents are told to ask a human.
Site: yarrow-fixes.vercel.app
More from coding & agent
- Sentdex: DeepSeek's Browser Harness Is 'Actually Quite Good' Running Locally — Sentdex · 2026-10-21
- 3 underrated OpenAI developer combos: voice agents, agent orchestration, and Codex integrations — gabrielchua · 2026-10-21
- GPT-6.1 Agent With Full Access Nukes Windows Bloatware, Sparks 'This Is AGI' Hype — keyanzhang · 2026-10-21
- MMEnterprise-Bench from KAIST Tests Enterprise Agents on Conflicting, Evolving Multimodal Records — Seohyun Lee · 2026-10-21
- NTU and Alibaba's FocusGUI: a lightweight adapter filters multimodal evidence to stop GUI agent error compounding — NanyangTechnologicalUniversity · 2026-10-21
- Snowflake's HybridResearcher: a 4B agent trained on synthetic cross-source trajectories beats much larger models — Snowflake · 2026-10-21