Abstract
Dispatching an autonomous coding subagent against a ticket — letting a large-language-model agent edit a repository, commit, and open a pull request with no human reviewing each action — requires governance that degrades safely rather than failing in either of the two unsafe directions: refusing outright (which stalls delivery) or running unbounded (which burns spend and ships defects). This publication discloses a chained, fail-closed precondition envelope of eight ordered locks evaluated before any run is permitted to touch the repository, together with a post-run settlement step. Three of the locks carry the inventive weight of the composition: (i) a per-principal autonomy flag guarded by a consecutive-failure circuit breaker that increments on each failed run, resets to zero on the first success, and auto-disables the principal at a configured failure threshold; (ii) a daily-budget check whose side effect is to defer the ticket (set a deferreduntil timestamp re-queuing it for the next day) instead of refusing it or silently downgrading the model; and (iii) a higher-autonomy tier that additionally requires a per-dispatch activation token before the envelope will admit the run. Every outcome — admit, defer, abort, fail, succeed — emits one per-run audit record. We present the architecture, the ordered-lock state machine, a relational data model, a worked dispatch scenario, a failure-mode analysis, framework mappings, a clean-room Node.js reference implementation reducing the composition to practice, and one independent plus fifteen dependent inventive aspects. The document is published as dated prior art to bar later patenting of the combination by others; the individual locks are acknowledged as known art.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Assuncao, gustavo matthew, "Eight-Lock Autonomous Dispatch Envelope (Self-Tripping Autonomy Breaker)", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10859