discipline

name	discipline
description	Run Holzmann "Power of Ten" discipline review on changed code. Checks nesting depth, loop bounds, function size, preconditions, mutable state, side effect surfacing, and indirection depth. Use after writing code, before committing, or as part of PR review. Outputs a weighted letter grade (A-F) with stoplight color.

Discipline Review

Enforce Holzmann's "Power of Ten" rules (adapted for F#) on changed code.

Arguments

• --scope <mode> — What to review. Options: diff (default), branch, pr, full, or a file/directory path.
• --voice <name> — Output persona. Options: mission-control (default), stephen-fry, neil-degrasse-tyson, david-attenborough, gordon-ramsay.

Scope Modes

Scope	Files reviewed	Resolved via	When to use
diff (default)	Unstaged + staged changes	git diff --name-only -- '*.fs' + git diff --cached --name-only -- '*.fs'	After writing code, before committing
branch	All changes since branching from main	git diff main...HEAD --name-only -- '*.fs'	Before creating a PR
pr	Files changed in current PR	gh pr diff --name-only filtered to *.fs	Reviewing an open PR
full	All source files in the project	find src/ test/ sample/ -name '*.fs'	Baseline audit, tech debt sweep
<path>	Specific file or directory	Direct path, glob if directory	Targeted review

For full scope on large codebases: warn the user this may take a while, then proceed.

Process

1. Parse arguments. Extract --scope (default: diff) and --voice (default: mission-control).

2. Resolve the file list based on scope:

   • diff: Run git diff --name-only -- '*.fs' and git diff --cached --name-only -- '*.fs'. Combine results.
   • branch: Run git diff main...HEAD --name-only -- '*.fs'.
   • pr: Run gh pr diff --name-only, filter to *.fs files.
   • full: Run find src/ test/ sample/ -name '*.fs' (adjust paths if project structure differs).
   • <path>: If a file, use directly. If a directory, find all *.fs files within it.

3. If no files found, ask the user which files to review.

4. Dispatch the code-discipline agent on the resolved file list. Include in the agent prompt: Voice: <voice-name> and Scope: <scope> so the agent has context.

5. The agent returns findings with a grade summary. Present it directly.

6. For Critical or High findings (automatic F), offer to fix them (with user approval).

Grading

Grade	Range	Stoplight	Meaning
A	93-100	🟢 Green	Ship it
A-	90-92	🟢 Green	Clean
B+	87-89	🟡 Yellow	Minor polish needed
B	83-86	🟡 Yellow	Acceptable
B-	80-82	🟡 Yellow	Review recommended
C+	77-79	🟠 Orange	Needs work
C	73-76	🟠 Orange	Significant issues
C-	70-72	🟠 Orange	Borderline
D	60-69	🟠 Orange	Do not merge
F	0-59	🔴 Red	Failing

Severity Weights

Severity	Rules	Per finding	Auto-fail?
Critical	13, 14	Score = 0	1 = instant F
High	9, 10, 12	Score = 50	1 = instant F
Medium	11, 15	-8	6 = F
Low	—	-3	14 = F

Rules Checked

#	Rule	Severity	What it catches
9	Keep It Linear	High	Nesting > 2 levels deep
10	Bound Every Loop	High	Unbounded recursion, retry, poll
11	One Function, One Job	Medium	Functions > 60 lines, "and" descriptions
12	State Your Assumptions	High	Missing precondition checks at boundaries
13	Narrow Your State	Critical	Module-level mutable, distant state
14	Surface Your Side Effects	Critical	I/O hidden behind pure-looking names
15	One Layer of Indirection	Medium	>1 layer of dispatch to reach actual work

When to Use

• After writing code, before committing
• As part of /expert-review (dispatch alongside other experts)
• When reviewing AI-generated code (which routinely violates these rules)
• When the user asks "does this follow the discipline rules?"