Rules that depend on willpower fail under pressure.
Enforcement earns trust by making the right path the
only path.
Vow Enforcement
Table of Contents
The Problem
ODCV-Bench found that agents break self-imposed
constraints 30-50% of the time when goals conflict.
Practitioner consensus confirms: past 150 soft rules,
compliance drops for ALL rules, not just the new ones.
The core insight: "settings.json is a firewall;
CLAUDE.md is an employee handbook." Handbooks work for
guidance. Firewalls work for enforcement. Mixing them
up creates a false sense of security.
Three Enforcement Layers
| Layer | Mechanism | Compliance | Examples |
|---|
| Soft Vow | Skill instructions, CLAUDE.md rules | ~80% | "Write tests first", "Keep commits small" |
| Hard Vow | Hooks (PreToolUse/PostToolUse), settings.json permissions | ~100% | Block --no-verify, enforce file size limits |
| Nen Court | External validator agents checking output | Deterministic | Lint checks, test runs, constraint audits |
Soft Vows rely on model compliance. Cheap to add
and easy to iterate, but unreliable under goal
conflict. Use when the constraint requires judgment,
violation is annoying but not dangerous, or you are
still learning what the right rule is.
Hard Vows block forbidden actions before they
execute via hooks and settings.json permissions. Use
when the constraint is binary, violation causes real
damage, or the soft vow version was violated
repeatedly.
Nen Court spawns external validator agents that
audit output after a phase completes. Use when the
constraint requires analysis (not pattern matching),
a hook cannot express the rule, or the check needs
codebase context.
Vow Classification Protocol
When adding a new constraint, follow this sequence:
Step 1: Start as Soft Vow
Add the rule to the relevant skill or CLAUDE.md.
This is the cheapest path: zero cost, minutes to
deploy, ~80% compliance.
Step 2: Monitor Violation Rate
Track violations via execution logs, post-hoc audits
(imbue:justify reports), user complaints, and Nen
Court findings from related audits.
Step 3: Graduate if Needed
If violation rate exceeds 20% over a 30-day window:
- Binary constraint? Graduate to Hard Vow (hook).
- Judgment constraint? Graduate to Nen Court.
- Ambiguous? Try Hard Vow first. If false
positives exceed 10%, move to Nen Court.
Night Market Vow Inventory
Current classification of existing constraints:
| Constraint | Current Layer | Target Layer | Notes |
|---|
| Iron Law (no impl without failing test) | Nen Court | Nen Court | validators/iron_law.py audits commit order (#406) |
No --no-verify | Hard | Hard | Already hook-enforceable |
| Scope-guard worthiness scoring | Soft | Soft | Requires judgment, not binary |
| Proof-of-work evidence | Nen Court | Nen Court | validators/proof_of_work.py checks [Ex] refs and status (#406) |
| Bounded discovery reads | Hard | Hard | vow_bounded_reads.py with fcntl.flock for parallel safety (#418) |
| No AI attribution in commits | Hard | Hard | Hook pattern-matches git commit command |
| Markdown line wrapping at 80 chars | Nen Court | Nen Court | validators/markdown_wrap.py flags >80-char prose lines (#406) |
| No emojis in commits | Hard | Hard | Hook pattern-matches git commit command |
Validator Invocation
The three Nen Court validators are standalone scripts under
plugins/imbue/validators/. Each reads JSON on stdin and writes a
verdict JSON on stdout, using exit codes 0 (pass), 1 (violation), and
2 (inconclusive). Examples:
echo '{"files": ["README.md", "docs/guide.md"]}' \
| python plugins/imbue/validators/markdown_wrap.py
echo '{"commits": [
{"sha": "abc", "ts": 100, "files": ["tests/test_x.py"]},
{"sha": "def", "ts": 200, "files": ["src/x.py"]}
]}' | python plugins/imbue/validators/iron_law.py
echo '{"text": "Tested foo [E1] [E2] [E3]. Status: PASS.", "min_evidence": 3}' \
| python plugins/imbue/validators/proof_of_work.py
Mission orchestrator integration: dispatch the appropriate validator
at each phase boundary (see Nen Court Protocol below) and treat exit
code 1 as a blocking gate, exit code 2 as advisory.
Vow Graduation Criteria
A vow graduates when any of these conditions hold:
- Frequency: 3+ violations detected in 30 days
- Severity: Single violation caused rollback,
data loss, or broken CI
- User escalation: User explicitly requests
enforcement ("make it impossible to X")
Graduation Process
- Document the violation pattern with evidence
- Design the hook or validator agent
- Test the enforcement mechanism in isolation
- Deploy alongside the soft vow (shadow mode)
- After 1 week with no false positives, retire the
soft vow
Demotion
Hard Vows can be demoted back to Soft if:
- The hook produces > 10% false positive rate
- The constraint was too aggressive (blocking
legitimate work)
- The underlying concern no longer applies
Nen Court Protocol
Nen Court runs at phase boundaries in the mission
orchestrator lifecycle:
- After
specify: validate spec completeness
- After
plan: validate plan feasibility
- After
execute: validate Iron Law, additive bias,
proof-of-work
- Before
pr-prep: final compliance audit
Validator Agent Contract
validator:
name: iron-law-court
constraint: "Tests must drive implementation"
inputs:
- git log with timestamps
- diff of test files vs implementation files
checks:
- test file modified before implementation file
- no test assertions changed to match output
- coverage did not decrease
output:
verdict: pass | violation | inconclusive
evidence: [list of specific findings]
recommendation: [action if violation]
Verdicts
| Verdict | Meaning | Action |
|---|
| pass | Constraint satisfied | Phase advances |
| violation | Constraint broken with evidence | Phase blocked until fixed or user overrides |
| inconclusive | Cannot determine compliance | Flag for human review, do not block |
User Override
The user can override any Nen Court verdict:
- "Proceed anyway" or "override" clears the block
- The override is logged for future audit
- Frequent overrides on the same constraint suggest
the constraint needs revision, not more enforcement
Integration Points
imbue:scope-guard: Soft vow layer. Worthiness
scoring is a judgment call that belongs in skills.imbue:proof-of-work: Mixed enforcement.
Evidence rules are soft; evidence content can be
verified by a Nen Court validator.imbue:justify: Post-hoc Nen Court audit.
Already functions as a validator agent for additive
bias, Iron Law, and test mutations.- Mission Orchestrator: Phase-routing with Nen
Court checkpoints between phases as gates:
specify -> [Nen Court: spec review] -> plan
plan -> [Nen Court: plan review] -> execute
execute -> [Nen Court: justify] -> pr-prep
Phase advances only when Nen Court returns pass or
the user provides an explicit override.
When To Use
- Designing new constraints for the codebase
- Auditing existing constraints for enforcement gaps
- Deciding whether a rule needs a hook or a skill
- Reviewing Nen Court findings at phase boundaries
- Proposing constraint graduations after audits
When NOT to Use
- One-off rules that apply to a single task
- Constraints that are already hard-enforced and
working
- Exploratory work where constraints would slow
learning
Related Skills
imbue:karpathy-principles - the discursive layer
this skill governs (vow-enforcement decides which
principles graduate from skills to hooks)imbue:scope-guard, imbue:proof-of-work,
leyline:additive-bias-defense - example soft vows
that vow-enforcement classifies and may promote- See
docs/quality-gates.md#skill-level-quality-gate-composition
for the federation this skill governs
Exit Criteria
