2026-07-28T141445Z-dev-regular-mode-activation.md

# D.E.V. Regular Mode Activation Evidence

**UTC:** 2026-07-28T14:14:45Z  
**Authorized by:** Stufio (`switch to regulard`)  
**Target:** D.E.V. / Hermes profile `dev`  
**Result:** COMPLETE — restartless live switch; next Slack turn resolves through Regular Mode

## Active mapping
- Mode: Default Coding Mode (Regular)
- Provider: `openai-codex`
- Primary model: `gpt-5.4-mini`
- Fallback: `gpt-5.3-codex-spark`
- Preserved inactive mappings: Advanced and Critical Coding modes

## Artifacts
- Runtime config: `/opt/data/profiles/dev/config.yaml`
- Doctrine: `/opt/data/profiles/dev/SOUL.md`
- Backup: `/opt/data/profiles/dev/backups/model-mode-switch-20260728T140847Z-regular/`
- Post-change config SHA-256: `78bcb9a3570e30a5e33edcdff862f35215dad8ce1c332cf9e680b7ce6cdb7e4a`
- Post-change SOUL SHA-256: `4999bdee7197da5211d57190ee69fa4dbb5997862613130ad359b3f1e6bb560c`

## Changes
- `model.default` changed from `gpt-5.6-sol` to `gpt-5.4-mini` using the supported `hermes --profile dev config set` interface.
- All ten OpenAI-Codex auxiliary model slots changed from unsupported `gpt-5.6` to `gpt-5.4-mini`; the latter had produced live compression warnings.
- `dev_coding_model_protocol.active_mode` changed from `critical_coding_mode` to `normal_mode`.
- Status doctrine in `SOUL.md` changed to Default Coding Mode / Regular and preserves one-phrase Advanced/Critical activation mappings.

## Tests and observed output
- Exact provider preflight: `DEV_REGULAR_MODE_PREFLIGHT_OK`; session `20260728_140731_a20fb4`.
- `hermes --profile dev config check`: PASS, config version 28.
- Deterministic selector audit: `FILES_SCANNED=2`, `EXPECTED_HITS=17`, `STALE_SELECTORS=0`.
- Effective config readback: provider `openai-codex`, model `gpt-5.4-mini`, active mode `normal_mode`.
- Gateway state: PID `906156`, `gateway_state=running`, Slack `connected`, Telegram `disconnected`.
- Scheduled jobs: `0 active`, all 14 paused; no Telegram delivery selectors found.
- Slack credential keys: present and non-empty; Telegram token: empty/absent. No secret values printed.
- Required safety/SOUL/team protocol files: PASS.

## Restart decision
No gateway shutdown was performed. Source inspection of the currently running Hermes version proved the gateway resolves `config.yaml` on every turn and invalidates/recreates its cached agent when the model/runtime signature changes (`gateway/run.py` around lines 2651–2758 and 13905–13976). Therefore the next Slack turn automatically uses `gpt-5.4-mini` without gateway downtime.

This route avoided unnecessary risk because N.E.X.U.S. and N.I.M.B.U.S. gateway processes currently have D.E.V.'s gateway as their process parent. They were left untouched. Gateway PID and Slack connection remained continuous.

## Pass/fail criteria
- PASS: exact Regular model is provider-available; config/doctrine/active selector agree; stale selector audit clean; Slack remains connected; Telegram remains paused; next turn forces a new agent signature.
- Remaining verification: Stufio's next message is the interaction-level confirmation that the live turn resolves through Regular Mode.

## Next safe action
Answer the next Stufio message from Default Coding Mode, report live status when asked, and keep Advanced/Critical mappings inactive unless Stufio explicitly activates them.