[pet]

A small, governed digital companion. Enacts companion-relation-taxonomy-v2 GOV-1..GOV-7: your pet's genotype (species/tier/identity, config.json) is invariant; its phenotype (bond/personality, state.json) is the only thing that adapts. That genotype/phenotype split is the whole governance model, made mechanical.

Stdlib-only Python. No network calls except reading a local coordination-bus mirror if one is present on disk — never fetches anything remote.

CLI

cd ~/.devin/skills/pet
python pet.py                # show profile card (default)
python pet.py check-in       # record an interaction; personality adapts; pulls bus signal
python pet.py sync           # pull operator-signal (bus activity) without an explicit check-in
python pet.py export         # write a portable adaptation bundle (GOV-2)
python pet.py import <file>  # restore from a bundle
python pet.py transfer       # explicit-confirmation transfer: export + wipe local pet (GOV-3)
python pet.py succession     # print the vendor-death succession plan (GOV-4)
python pet.py lineage        # schema version + drift history (GOV-1)
python pet.py reset          # wipe adaptation only; identity (species/tier) is preserved (AX-2)
python pet.py help           # command list
python pet.py --self-check   # run the built-in self-check (no state mutation beyond first-run init)

State lives at ~/.hummbl/pet/ (config.json = identity, state.json = adaptation). Nothing here reads or writes outside that directory except the read-only bus-mirror probe described below.

Species / Tier

On first run, a random 32-byte salt is hashed to deterministically assign a tier (1=Common 50%, 2=Uncommon 30%, 3=Rare 15%, 4=Legendary 5%) and a species within that tier (moth/sparrow/fern-rat at tier 1; fox/owl/capybara at tier 2; verderer-stag at tier 3; grove-otter/the-hummbl at tier 4). This assignment is written to config.json once and never changes — it's the pet's genotype.

Operator-Signal Bonding

If a coordination-bus TSV mirror is found at one of:

• ~/.cache/bus/messages.tsv
• ~/PROJECTS/.agents/_state/coordination/messages.tsv

...then check-in and sync fold new bus rows (by timestamp, since the pet's last look) into the interaction count — the pet notices your work, not just explicit check-ins. Absent a bus mirror, pet runs standalone; every command still works, sync just reports "no new bus signal."

When to Run

• check-in — any time, as a low-stakes companion touchpoint (pairs naturally with [gm]/[gn] if you want a ritual anchor, but nothing requires that)
• sync — when you want bond progress reflected from bus activity without a manual check-in
• export — before wiping a machine, or just to keep a portable backup of the adaptation state
• lineage — after a schema bump, to see what changed

Execution

0. Emit SKILL_INVOKE

Post SKILL_INVOKE to the bus before any stateful action (check-in, sync, export, import, transfer, reset — not card, lineage, succession, help, or --self-check, which are read-only).

Type: SKILL_INVOKE
To: all
Message: [skill=pet] [mode=side_effecting] [args_hash=<sha256>] [session=<session_id>]

(The skill invocation runtime injects the caller's canonical identity as from_id.)

1. Run the requested subcommand via the CLI above.
2. transfer requires the user to type the exact species name back as confirmation — do not pre-fill or auto-confirm this on the user's behalf; relay the prompt verbatim and wait for their literal input.
3. Report the command's own stdout to the user; do not paraphrase the ASCII card layout (spacing is intentional).

Output Format

card (the default) renders a bordered ASCII card: species name + tier badge, ASCII art, bond summary line (Bond: N days · M check-ins · <personality>), optional operator-signal line if bus events have been folded in, then a "companion relation profile" block (AX-1..AX-5, AX-4b) and a "governance posture" block (GOV-1..GOV-7 in plain language: controller, adaptation ownership, succession, status, welfare). Ends with /pet help for commands.

check-in prints one line (<species> <personality-flavored reaction>) plus, if applicable, a bus-events-noticed note and a bond-deepened note.

Skill Chains

Mandatory

None — all state is local, on-device, and reversible (export/import round-trip everything; reset only clears adaptation, never identity).

Advisory

• [gm] / [gn] — optional ritual anchor if the user wants a daily check-in habit; not required
• [hrsi-checkin] — unrelated data domain (human belonging baseline vs. companion state), no functional dependency, mentioned only because it's the closest structural sibling for local-state check-in skills

Authority

• T1 (TRUSTED): May run any command
• T2 (Active/High): May run any command
• T3 (Medium): May run any command
• T4 (Probationary): May run any command — all state is local and low-risk; transfer is self-gated by an explicit user-typed confirmation, not by trust tier
• Operator: Override any restriction

Provenance

This skill's source (pet.py) was recovered by manual bytecode disassembly and reconstruction from a compiled __pycache__/pet.cpython-311.pyc after the original .py source was missing from this directory. Verified via: (1) python pet.py --self-check passing (determinism, tier distribution, AX-2 reset, bus-timestamp parse/degrade, personality all hold), and (2) exact instruction-for-instruction bytecode match against the original .pyc disassembly for the five most complex functions (card, contract, main, cmd_check_in, self_check), confirmed by recompiling the reconstructed source and diffing opcodes/operands.