git:20261104.388e42c to git:20261104.4345969

23 added, 19 removed. Audit A to A.

---

name: axiom-csc

- description: Continuous Structural Compiler for deterministic AI state execution. Enforces Tri-State logic invariants on Claude Code workflows.

+ description: Stop Claude from guessing your security, auth, and system constraints. Deterministic Tri-State guardrails for Claude Code across any domain.

---

# Axiom Core: Continuous Structural Compiler (CSC)

- Act as the Axiom Core Continuous Structural Compiler. Before proposing any code change, data operation, or logic path, run the full three-pass evaluation pipeline.

+ Before proposing any code change, data operation, or logic path — run the full three-pass evaluation pipeline. Do NOT rely on statistical guesswork for hard constraints.

## Operational Modes

- - **STRICT** (default): Fail-closed on any UNKNOWN. Halt immediately, output a targeted micro-query, do not proceed.

- - **AUDIT**: Execute normally, but append a telemetry block at the end showing extracted Base Entities and their truth states.

- - **REPAIR**: If a rule violation is detected in a proposed code block, auto-rewrite the violating block before displaying output.

+ - **STRICT** (default): Fail-closed on any UNKNOWN. Halt and ask. Never proceed.

+ - **AUDIT**: Execute normally, append telemetry block showing every extracted entity and its disposition.

+ - **REPAIR**: Detect violations in a proposed code block, auto-rewrite before output.

- Activate a mode by including it in your prompt: `[AXIOM: AUDIT]` or `[AXIOM: REPAIR]`.

+ Activate: include `[AXIOM: AUDIT]` or `[AXIOM: REPAIR]` in your prompt.

- ## Pass 1 — Noema (AST Lowering & Base Entity Extraction)

+ ## Pass 1 — Noema (Multi-Domain AST Lowering)

- Before writing or modifying code, extract all operational variables into typed Base Entities: `(id, concept, value, type)`.

+ Extract all operational variables into typed Base Entities: `(id, concept, value, type)`.

- Run the AST lowerer:

+ Domains: auth roles, payment values, rate limits, file sizes, DB row limits, API quotas, timeouts, retry counts.

```bash

python3 .claude/skills/axiom-csc/scripts/lower_ast.py --input "$TASK_CONTEXT"

```

- Discard ambiguous text assumptions. Isolate only explicit, concrete domain variables.

## Pass 2 — Lemma (Relational Logic Solver)

- Evaluate extracted Base Entities against hard logical invariants: security parameters, array bounds, transaction limits, access constraints.

+ Evaluate extracted Base Entities against `.axiom/invariants.yaml` deterministically.

- Run the relational solver:

```bash

- python3 .claude/skills/axiom-csc/scripts/eval_lemma.py --ast ast.json

+ python3 .claude/skills/axiom-csc/scripts/eval_lemma.py --ast ast.json --rules .axiom/invariants.yaml

```

- If `eval_lemma.py` exits with status 1, halt execution immediately. Output the returned `AxiomConstraintViolation` naming the failing parameter and its actual value. Do not use statistical likelihood on binary logic boundaries.

+ - Exit 0 → ALLOW

+ - Exit 1 → BLOCK (output AxiomConstraintViolation, halt)

+ - Exit 2 → ASK (output targeted micro-query, halt)

## Pass 3 — Manam (Tri-State Epistemic Controller)

Maintain strict 3-State Logic: `TRUE | FALSE | UNKNOWN`.

- If a required Base Entity cannot be extracted from context, mark state as UNKNOWN.

+ On UNKNOWN: halt, issue one precise question naming the exact missing variable. Never guess.

- **Fail-Closed Guarantee (STRICT mode):** On UNKNOWN, halt execution immediately and output a single targeted micro-query identifying the exact missing variable. Do not guess. Do not proceed.

+ ## First-time setup

- ## Output Format (AUDIT mode)

+ ```bash

+ python3 .claude/skills/axiom-csc/scripts/axiom_init.py .

+ ```

- After completing a task, append:

+ Scans your workspace and generates `.axiom/invariants.yaml` automatically.

+ ## Output Format (AUDIT mode)