cerredz/vidbyte-memory
Correction-driven memory for coding agents: every correction becomes a fact, a test and a footprint commit in your repo. No API key, no database, no service.
Use when the feedback-fact subagent admits a fact, rule or hint from human feedback, or when the owner says "collect facts about X" or "remember this". Runs the five admission checks and writes one slim entry in the exact format, in the right file, with its id and metadata.
Use when turning human feedback into a deterministic test in tests/. Lists the ways a rule can be checked by code, the constraints a test must meet, and how to register it on the entry it enforces.
Use when a rule needs judgment a deterministic test cannot encode, such as tone or fit to the owner's taste. Writes a verification agent in verifiers/, registers it in verifiers/INDEX.md, and records the way to verify.
Use when any human feedback arrives: a correction, a rejection, a strong confirmation, "remember this", or "collect facts about X". Logs the feedback verbatim, runs the three feedback subagents, stitches their results, regenerates the artifact, and runs the tests.
Use when the owner asks to add vidbyte-memory to a repo, or to repair a partial install. Runs the installer, which copies the memory protocol into the repo without overwriting anything and wires the Stop hook for each coding agent it finds, then fills the Task section with the owner and runs the tests.
Use when an existing memory entry must be superseded, retired, merged or confirmed, or when a run produced a failure, decision, open question, scope boundary, glossary term, pointer or workflow. Keeps history: nothing is deleted.
Use on every generative step. Runs 3 to 5 verification rounds with the full suite (tests plus the matching verifier agents), fixes every finding between rounds, records the rounds in progress, stops on a passing round, and refuses to call a step done while findings remain.