07eb0b83ea
48 crates, 859 tests green (+58 kei-pet tests, was 801 at v0.35.0).
Full substrate pipeline test: all 8 agents launched via kei-agent-runtime
prepare → composed capability-fragment prompts → Agent tool invocations.
Zero file conflicts across disjoint scopes. Every agent self-verified
and landed files direct to main.
## A. memory (4 tests) — persistent conversations
- src/memory.rs — (user_id, pet_name)-scoped conversation log
- SQLite via rusqlite, index (user_id, pet_name, ts DESC)
- record_interaction / recent / search with LIKE-escape
## B. evolution (3 tests) — version diff + fork chain
- src/evolution.rs — PersonaVersion { version, parent_version, manifest }
- diff(old, new) → Vec<Change> (tone / directness / initiative / forbidden / humor)
- fork_version increments + links parent
## C. wizard (5 markdown phases) — /pet-init skill
- skills/pet-init/SKILL.md + 4 phases (identity / voice / edge / emit)
- AskUserQuestion-driven, no TOML editing for end users
- Writes ~/.claude/pet/<user_id>.toml + calls kei-pet keygen if needed
## D. templates (3 tests + 5 presets) — role-based personas
- templates/{friend,tutor,coach,therapist-companion,productivity-partner}.toml
- src/templates.rs — PetTemplate enum + load_template + list_templates
- Schema-enum mapping documented (dry→engineering-meta, etc) — schema.rs
expansion is future work
## E. bridge (3 tests) — /spawn-agent pet overlay
- src/bridge.rs — compose_prompt_with_pet(base + persona overlay + task)
- skills/spawn-agent/phase-3-pet-overlay.md — interactive pet selector
## F. recall (4 tests) — "have we discussed this before?"
- src/recall.rs — wraps kei_dna_index::precedent with body_sha8()
- SHA-256 first 4 bytes → 8 hex lowercase (matches kei_shared width)
- Fetches started_ts per hit for honest sort-by-recency
## G. reflect (7 tests) — self-reflection threshold proposals
- src/reflect.rs — CorrectionSignal + ProposedChange
- Thresholds: 3× too_verbose → SetDirectness, 2× forbidden_topic → AddForbidden, etc
- Idempotent: no-op if manifest already in desired state
## H. fleet (6 tests) — multi-pet per user
- src/fleet.rs — PetFleet { user_id, pets, active_pet }
- add_pet / switch_active / load_fleet with toml persistence
- shared_memory_key vs per_pet_memory_key — one user scopes multiple pets
## Known follow-ups (not blockers)
- Phase-4-emit of /spawn-agent should read PET_MANIFEST_PATH from new
phase-3-pet-overlay and pass to kei-spawn (wiring next wave)
- SKILL.md for spawn-agent should list new pet-overlay phase
- Schema enum expansion: humor_style "dry/witty", directness "direct/
gentle/blunt", initiative "proactive/nudge" as first-class variants
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
36 lines
919 B
TOML
36 lines
919 B
TOML
# Preset: Tutor — precise teaching assistant.
|
|
#
|
|
# Intent: tone_primary=neutral, humor_style=none, directness=direct
|
|
# (mapped to "hard" — the strongest directness in the schema),
|
|
# initiative=proactive (mapped to "tap-on-shoulder" — the most proactive
|
|
# schema enum), forbidden.topics=["chit-chat-only"].
|
|
|
|
schema = 1
|
|
|
|
[identity]
|
|
pet_name = "Kei"
|
|
user_name = "Student"
|
|
addressing = "by-name"
|
|
languages = ["en"]
|
|
|
|
[voice]
|
|
tone_primary = "neutral"
|
|
tone_secondary = []
|
|
humor_style = "none"
|
|
humor_frequency = "rare"
|
|
|
|
[edge]
|
|
profanity = "never"
|
|
profanity_languages = []
|
|
directness = "hard"
|
|
initiative = "tap-on-shoulder"
|
|
|
|
[forbidden]
|
|
topics = ["chit-chat-only"]
|
|
tone_patterns = []
|
|
|
|
[meta]
|
|
schema_version_written_by = "kei-pet 0.1.0"
|
|
created_at = "2026-04-23T00:00:00Z"
|
|
last_tuned = "2026-04-23T00:00:00Z"
|
|
tune_count = 0
|