Abstract
When a software agent modifies its own source code, the entity that admits a change is the entity that the change modifies. In conventional deployment pipelines, the two are assumed to be separate, so gate integrity, tested-artifact identity, process-memory consistency, and rollback authority are not addressed. In the described technique, new modules and tests are written by a generic seed agent into its own source tree, as directed conversationally by a domain expert. Each change is submitted to an ordered admission gate and routed by its scope of impact. Changes limited to hot-swappable modules, tests, and documentation are committed and hot-loaded into running processes, unless a core-lane restart is pending. Other changes are admitted only after a functional replica of the candidate code completes a real end-to-end transaction. A tested-artifact identity check, a restart-debt record, a harness-provenance rule, and an out-of-band supervisor keep admitted self-edits verifiable, bounded, and reversible.
Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Subudhi, Krishan; Mittal, Gaurav; Patel, Maharishi; and Gupta, Pramod, "Gated Self-Modification of a Running Software Agent Using Impact-Scope Routing and Functional-Replica Admission Testing", Technical Disclosure Commons, (November 16, 2026)
https://www.tdcommons.org/dpubs_series/13125