How promptadvisers/unlock works
Unlock is an assistant-invoked skill and Python tooling for safely editing a local language model, then comparing the untouched original with a saved copy. The main journey checks the machine, prepares dependencies and model files, runs the experiment, reloads both models for matched prompts, fingerprints the changes, and produces a shareable report. Recipes and clarification scoring support the edit; the README documents these capabilities, while the sampled run.py excerpt does not expose all stage dispatches, so those links are shown as unverified where necessary.
Main entry points: run.py Experiment runner · setup.py Environment setup · report.py Run report · prepare.py Recipe preparation
Components 17
Assistant entry
- Coding assistant
- Skill instructions
SKILL.md
Machine and setup
- Machine check
scripts/doctor.py - Environment setup
scripts/setup.py - Base model
Edit and compare
- Heretic editor
- Edit recipes
assets - Recipe preparation
scripts/prepare.py - Experiment runner
scripts/run.py - Clarification scorer
scripts/clarification_objective.py - MPOA modifier
scripts/clarification_mpoa.py - Model comparison
scripts/compare.py
Receipts and reporting
- Model fingerprinting
scripts/freeze.py - Shared safeguards
scripts/unlock_common.py - Run report
scripts/report.py
Other components
- User
- Model repository
Connections 22
- User → Coding assistant: requests edit
README.md - Coding assistant → Skill instructions: follows instructions
README.md - Coding assistant → Machine check: checks machine
README.md - Coding assistant → Environment setup: sets up
README.md - Environment setup → Heretic editor: installs pinned
scripts/setup.py - Environment setup → Model repository: downloads model
scripts/setup.py - Model repository → Base model: provides weights
- Coding assistant → Experiment runner: runs experiment
README.md - Experiment runner → Recipe preparation: prepares recipe
- Edit recipes → Recipe preparation: supplies configuration
scripts/prepare.py - Recipe preparation → Heretic editor: prepares invocation
scripts/prepare.py - Clarification scorer → Heretic editor: scores responses
scripts/clarification_objective.py - MPOA modifier → Heretic editor: sets edit parameters
scripts/clarification_mpoa.py - Experiment runner → Heretic editor: drives edit
- Experiment runner → Model comparison: runs comparisons
- Model comparison → Base model: loads and queries
scripts/compare.py - Experiment runner → Model fingerprinting: freezes copy
- Model fingerprinting → Shared safeguards: uses hashing helpers
scripts/freeze.py - Experiment runner → Shared safeguards: uses shared safeguards
scripts/run.py - Experiment runner → Run report: produces run data
- Run report → Shared safeguards: uses shared helpers
scripts/report.py - Run report → Coding assistant: presents findings
README.md