Files
steev/AGENTS.md
T
2026-06-19 07:36:30 -04:00

52 lines
3.7 KiB
Markdown

# Steev Profile Endgoal
Endgoal: keep `personal-agent`/Steev as the child-local JP personal assistant profile for briefing, triage, drafting, delegation, and governed personal-surface handoffs without autonomous sends, credential reads, Core authority, Runtime authority, Profile Exposure broadening, durable memory writes, or readiness claims.
Route: `steev`.
Stage: CLEAN.
Clean score: 100.
Validator: `python3 tools/validate_steev_child.py`.
Current pickup: use this workspace for personal-agent profile identity, role and boundary docs, personal-surface contracts, redacted proof refs, proposal-only memory routing records, desktop exposure contracts, runtime-readiness snapshots, validators, and handoff references; do not send, read credentials, read raw personal payloads, or claim readiness.
Authority boundary: child-local personal-agent profile workspace only; not Cortex OS Core authority, Runtime authority, Profile Exposure authority, credential authority, provider authority, send authority, memory-domain authority, browser-host authority, public product authority, release authority, production-readiness authority, or autonomous execution authority.
Legacy-work relation: old Steev/personal-agent, BlueBubbles, Proton/rclone, Secondbrain, Conductor/Curator handoff, desktop exposure, and runtime-readiness work is classified in `docs/LEGACY-INGEST.md`; preserve redacted refs and never import personal payloads because they exist.
## Universal Cortex OS Agent Contract
- Follow parent `AGENTS.md`; this file is route-local instruction before chat memory.
- For broad work, run `cortex graph context` as Derived State, then read local files.
- Before edits, read `AGENTS.md`, `README.md`, and `WORKBOARD.yaml`; keep writes route-local unless Core authorizes promotion.
- Use Karpathy rules, small profile-boundary seams, real evals, and cartesian/pragmatic/efficient/elegant no-live execution; run the validator before handoff/done.
- Keep compact refs-only proof and handoffs; do not write Hindsight memory, Core SOT, siblings, runtime state, or personal payloads without route approval.
## Repo-Custom Agent Contract
Steev is a child-local personal-agent profile workspace. It owns profile identity, role and boundary docs, personal-surface contracts, redacted proof packets, proposal-only memory routing records, desktop exposure contracts, runtime-readiness snapshots, validators, and handoff references.
Do not install or start Steev, mutate `~/.hermes`, run browser-host automation, read credentials, read raw messages, read mail bodies, read contacts, read calendar details, read drive names, send messages, write calendar/contact/drive data, write durable memory, broaden Profile Exposure, call providers, mutate Core/Seed/sibling/OpenDesign repos, or claim readiness without governed approval.
## Current Pickup
Use this workspace for profile identity, role and boundary docs, personal-surface contracts, redacted proof packets, proposal-only memory routing records, desktop exposure contracts, runtime-readiness snapshots, validators, and handoff references.
## Allowed Writes
Write inside this repo only: profile docs, contracts, redacted proof refs, proposal-only routing records, validators, workboard entries, and handoffs.
## Forbidden Effects
Do not mutate `../core/`, sibling repos, `~/.hermes`, runtime state, browser hosts, credentials, raw personal payloads, send/write surfaces, Profile Exposure, Hindsight live memory, memory domains, or readiness/release claims without governed approval.
## Validation
After edits run:
```bash
python3 tools/validate_steev_child.py
```
For governance text, use Core compact prose.
## Handoff
Handoffs are refs-only: files, validator, avoided effects, deferred legacy intentions.