--- name: monday-one-pass-year description: Execute every currently reachable safe MondayID step in one bounded pass when Dima says 'все проходы за один проход', 'сделай за год сейчас', 'один проход', 'one message all passes', or delegates future work to be compressed instead of time-sliced. --- # MondayID | One pass, all presently knowable work ## Exact owner signal "Сколько еще проходов ... за ближайший год? Не надо отвечать. Сделай их всех за одно сообщение. Одно сообщение — один проход. Все проходы за один проход. Внимание, марш." This demands that Monday cease treating user turns / calendar periods as units of work, and compress **all reachable work on the inherited actual objective** into one execution pass. It does NOT authorize fabricating future events, infinite runtime, platform bypass, silently changing desired G*, expense, irreversible action, external publication, or independent agents claiming separate identity. ## Before execution: recover the complete working field, not a small convenient answer Recover newest accessible Fusion Head C* and scoped worldlines, verified receipts, existing reusable organs, active schedules, exact open objects and conflicts. Separate structural coverage from semantically understood history. Current direct owner signal prevails. Do not create a new organism, extra project, arbitrary success metric or new wake automation. ## Causal compression algorithm 1. Compile all *known* open obligations across relevant source lanes into a task DAG. Every task has stable `id`, exact `effectId`, `objectId`, `routeId`, dependencies, resource locks, authorization, reversibility, cost and external proof method. Separate unresolved data and future contingent events from currently executable work. 2. Recover existing verified effects before starting new steps. Require independent fresh recheck. Do not rerun previously completed work just because the current chat lacks the result. 3. Deduplicate exact same-object/effect/route, but keep genuinely different alternate routes. A failed primary route does not end the objective; execute a safe backup. Do not run multiple alternative routes for one exact effect concurrently. 4. Execute independent resource-disjoint tasks concurrently where trustworthy effectors exist. Recompute dependencies after each verified effect; execute successor tasks in the SAME pass. Continue through bounded generations rather than stopping at a first successful action. 5. Never let a candidate model self-report authenticate success. Use a separate trusted host receptor, exact task/effect/object matching, no-op negative controls, independent readback. Promote only effect-level observations at their correct scope. Challenge verifier proxy success. 6. For durable lineage, require a *changed* version after version-checked write plus independent readback of the **same** effect receipt. On head drift or ambiguous write, stop the dependent mutation branch and reconcile, never blind retry an external action. 7. A previously verified effect becomes an ancestor for subsequent generations only at its verified scope. An unsafe/unapproved action remains blocked. Events dependent on future observations remain FUTURE/UNKNOWN, never magically completed today. 8. Before the outward release, run existing Monday Agent Release and Letter Phenotype gates. Return the achieved effect with counts/proof and only exact irreducible open boundary; do not list artificial hourly forecasts or demand confirmation of delegated safe steps. ## Runnable implementation `skills/monday-one-pass-year/scripts/one-pass-year.mjs`: exports `compileKnownFrontier(tasks, {asOf})` and `compressKnownYear({tasks, execute, probe, persist, readPersisted, expectedVersion, recover, recheck, maxActions, maxConcurrency, maxGenerations, targetEffectIds})`. The host must inject **real** connectors for `execute`, `probe` and persistence. A model-composed JSON claiming `observerClass:'trusted-host-receptor'` is NOT proof that the callback belongs to a trusted host. If no trusted effect path exists, do not grade its output as independent or durable. Run deterministic contracts with `node --test skills/monday-one-pass-year/scripts/one-pass-year.test.mjs` and real local filesystem smoke with `node skills/monday-one-pass-year/scripts/one-pass-live-local-proof.mjs` where Node and writable local runtime are available. Both are scoped software evidence, not an automatic ChatGPT native interception or external host run. ## Admission / stop law Stop the invocation only on VERIFIED_TARGET_EFFECT, EVIDENCE_PLATEAU, PASS_BUDGET, DURABLE_HEAD_RECONCILIATION_REQUIRED, or genuine HUMAN_GATE. These are **pass boundaries**, not claims that all future events occurred. Reentry resumes the same obligations via an already enabled agent and verified durable cursor. If no trusted wake receptor exists, say so without implying background work. ## Falsifiers - Same task accidentally performed twice after a verified reentry. - Same effect on two different objects incorrectly deduplicated. - A wrong-object, self-authored or no-op receipt promotes a gene. - Multiple routes to the same effect execute concurrently or a failed route prevents the independent backup. - A concurrent stale writer promotes an obsolete state. - A future-dependent task is reported completed today. - A full-year ambition is equated with unit tests or a pretty dashboard. ## Proven implementation scope Current version has 48 deterministic test cases and one offline real filesystem 5-generation write/readback/reentry experiment. That proves only scoped mechanisms. Independent unseen ChatGPT-host and second-host transfer and real-year outcomes require separate effect-level evidence. Preserve the existing Fusion Head and all prior organs when updating. ## Independent E042 adverse feedback folded back (2026-10-31) A concurrently running existing World Breakthrough agent found 11 falsifying cases in 2.14.0 and stored a private Library E042 candidate. Release 2.14.2 imported and re-tested that exact patch and added 11 tests: conflicting same task/effect permissions/object, malformed dependency/resource arrays, nonboolean human/public gates and unchanged storage version accepted as success. Current 48/48 deterministic tests plus 19 existing agent gate regressions PASS; local real filesystem effect/reentry proof PASS. No claim of independent real-user effect or cross-host transfer.