What problem does it solve? Agents repeatedly waste time re-attempting approaches that already failed in a repository because there is no durable, curated record of what was tried and why it failed. ## Core Features & Use Cases - Dead-end ledger: Records approaches with a verdict (failed, abandoned, worked, partial), a topic, a reason, and an optional reference into .tried/ledger.md, committed alongside the code. - Cheap search before work: tried search matches queries against approach, reason, and topic so an agent can check for known dead ends before spending time. - Idempotent, concurrent-safe writes: Entries are content-hashed so identical records are not duplicated, and appends use a single O_APPEND write so parallel agents do not corrupt the file. - Use Case: Before re-architecting token refresh logic, run tried search "rotate refresh tokens" and discover a prior entry explaining the replay-window race that made it fail. ## Quick Start Ask the agent to check the tried ledger for previous attempts at your planned approach before starting work on it.