session close: PyPI on TODO, all v4.0 items closed, session-close-2 filed

Author: Douglas Jones

.dockerignore

@@ -0,0 +1,26 @@
+# Don't include development artifacts in the Docker image
+.git
+.gitignore
+.pytest_cache
+.playwright-mcp
+__pycache__
+*.pyc
+*.pyo
+*.egg-info
+dist/
+build/
+target/
+tests/
+examples/
+docs/
+dispatches/
+sessions/
+scripts/
+crates/
+*.cod
+*.png
+*.jpeg
+*.svg
+Cargo.toml
+Cargo.lock
+fly.toml

.kiro/specs/v4-language/G0_PROBLEM_STATEMENT.md

@@ -0,0 +1,141 @@
+# Codifide v4.0 — G0 Problem Statement
+
+**Date:** 2026-05-14  
+**Author:** Douglas Jones + Claude (Aegis/Harper)  
+**Status:** G0 — approved, proceeding to G1
+
+---
+
+## Why v4.0?
+
+The gap analysis from the "is this usable in the wild?" question identified
+four structural gaps between Codifide as a research prototype and Codifide
+as a usable tool. Each gap is a concrete, fixable problem. None requires
+rethinking the language design.
+
+---
+
+## Problem 1 — `sig` declarations are decorative, not enforced
+
+**What breaks without this:** An agent can declare `sig (n: Int) -> String`
+and pass a `Float` or a `List`. The runtime either silently coerces, fails
+at the primitive level with a confusing error, or produces wrong output.
+The type system is a lie. Agents that trust it will be misled.
+
+**Evidence:** Every case study agent wrote `sig` declarations. None of them
+were checked. The language claims to be designed for agents who need
+trustworthy contracts — but the most basic contract (type) is not enforced.
+
+**Scope:** Runtime type checking at call boundaries. Not full static type
+inference. Check argument types against `sig` declarations when types are
+known. Raise a typed `TypeViolation` error on mismatch.
+
+**Risk:** Medium. Touches the interpreter call path. Existing programs that
+accidentally pass wrong types will now fail loudly instead of silently.
+That is the correct behavior.
+
+---
+
+## Problem 2 — No standard library
+
+**What breaks without this:** Agents cannot write programs that read files,
+make HTTP requests, parse JSON, or do date arithmetic. Every real-world
+agent pipeline needs at least one of these. The current primitive set covers
+string manipulation and arithmetic but nothing that touches the outside world
+beyond `io.say` and `clock.now`.
+
+**Evidence:** The "usable in the wild" assessment identified this as a
+hard blocker for real-world use. The content-moderation pipeline task spec
+(the canonical test) uses only string primitives — it was designed to avoid
+this gap, not to demonstrate the language is complete.
+
+**Scope:** Four new effect groups:
+- `io.read` — read a file by path, return string
+- `http.get` / `http.post` — HTTP client primitives
+- `json.parse` / `json.encode` — JSON round-trip
+- `clock.date` — structured date arithmetic beyond `clock.now.hm`
+
+**Risk:** Medium. New effect declarations, new primitives, new error kinds.
+Does not touch existing primitives or the canonical form for existing programs.
+
+---
+
+## Problem 3 — No operated public registry
+
+**What breaks without this:** V3-2 shipped remote symbol resolution
+infrastructure, but the registry at `codifide.com/symbols/<hash>` is empty.
+Two agents cannot exchange symbols without out-of-band coordination because
+there is nowhere to publish to. The multi-agent protocol story is
+infrastructure without content.
+
+**Evidence:** V3-2 acceptance criterion was "agent on machine B resolves a
+symbol published by agent on machine A." That works mechanically but requires
+both agents to be running their own servers. A public registry with real
+symbols in it is the missing piece.
+
+**Scope:** Operate the public registry endpoint. Seed it with the canonical
+pipeline task spec symbols (the five programs from the case studies). Document
+the publish workflow. Add `codifide store push --registry https://codifide.com`
+as the canonical publish path.
+
+**Risk:** Low for the code (already exists). Medium for operations (requires
+a running server, storage, and uptime commitment).
+
+---
+
+## Problem 4 — Server is 127.0.0.1 only
+
+**What breaks without this:** The RPC server cannot be used for multi-machine
+agent coordination without a reverse proxy, TLS, and auth — none of which are
+documented or provided. The V3-2 remote registry works around this by using
+the public endpoint, but any team wanting to run a private registry is on
+their own.
+
+**Evidence:** The Sable audit of V2-1 flagged AUD-RPC-02 (no socket timeout,
+slow-loris risk) as P2. The server was explicitly documented as "local-only,
+trusted caller." That is the right call for v2.0 but not for v4.0 where
+multi-machine use is the goal.
+
+**Scope:** Add an `--auth-token` flag for bearer token authentication. Add
+TLS support via `--cert` / `--key` flags (or document the reverse proxy
+pattern). Remove the "not safe to expose over a network" warning when auth
+is configured. Update `docs/RPC_API.md`.
+
+**Risk:** Medium-high. Security-sensitive. Requires Sentinel review and
+Sable audit before shipping.
+
+---
+
+## Prioritization
+
+| ID | Problem | Priority | Risk | Dependency |
+|----|---------|----------|------|------------|
+| V4-1 | Runtime type enforcement | P1 | Medium | None |
+| V4-2 | Standard library | P1 | Medium | None |
+| V4-3 | Public registry (operated) | P2 | Low/Medium | V3-2 (shipped) |
+| V4-4 | Network-safe server | P3 | Medium-High | V4-3 |
+
+V4-1 and V4-2 are independent and can be implemented in parallel.
+V4-3 is mostly operational, not code. V4-4 depends on V4-3 being
+useful first.
+
+---
+
+## What is explicitly out of scope for v4.0
+
+- **Full static type inference** — V4-1 is runtime checking only. Static
+  inference requires a type system design that is not yet specified.
+- **Hosted runtime / cloud execution** — no adoption evidence.
+- **Time-indexed types (V3-4)** — still deferred, still no evidence.
+- **Editor integration** — still deferred.
+- **Structural diff and merge** — still deferred.
+
+---
+
+## G0 decision
+
+**Approved.** All four problems are real, bounded, and worth solving.
+Evidence is direct (case study findings, Sable audit findings, gap analysis).
+Scope is honest. Proceeding to G1.
+
+*Aegis sign-off: approved 2026-05-14*