A skill for Grok Build and Hermes Agent.
Long coding sessions with Grok keeps dying.
A tiny skill. It writes the plan as a numbered list, does one step, runs a real test, writes down the next step. Next chat reads that file and keeps going.
flowchart LR
A[activate] --> B[numbered plan]
B --> C[one step]
C --> D[real test]
D -->|not yet| C
D -->|green| E[write the next step]
E --> C
| File | Job |
|---|---|
.glrp/GOAL.md |
The whole plan, numbered. |
.glrp/UNIT.md |
The one step happening now. |
.glrp/progress.txt |
Steps already done. |
.glrp/config.toml |
verify — your real test command. |
Scripts: activate.py · next.py · check.py · close_unit.py
git clone https://github.com/msinclair25/glrp.git
bash glrp/install.shPuts the skill in ~/.grok/skills/glrp/ and ~/.hermes/skills/glrp/ when those folders exist. On Grok Build it also installs a Stop hook that kicks once if the model quits with steps left. Start a new session so it loads.
In a git project:
- Say
activate glrpor/glrp. - Set
verify = "pytest -q"(or whatever actually tests your project). - Paste the brief. It numbers the plan and starts step one.
- It does that step, runs the test, writes the next step, and keeps going this sitting.
- If you start a new chat, it reads the file and continues.
skill/scripts/ activate.py next.py check.py close_unit.py
.glrp/ GOAL.md UNIT.md progress.txt config.toml
python3 -m unittest discover -s tests -vMIT · Morgan Sinclair (@msinclair25)
