Self review functionality

.github/workflows/ci.yml

@@ -28,6 +28,15 @@ jobs:
         with:
           python-version: ${{ matrix.python }}
 
+      # Installed for every harness so a harness can adopt uv without a CI change.
+      # Cache disabled: harnesses that don't invoke uv yet leave no cache dir to save,
+      # which fails the post-job cache-save step.
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          python-version: ${{ matrix.python }}
+          enable-cache: false
+
       # Fail fast on malformed DAG / profile / benchmark JSON (the edits most likely to break a run).
       - name: Validate workflow & config JSON
         if: matrix.harness == 'security-harness'

.gitignore

@@ -41,3 +41,7 @@
 /security-harness/manifests
 /site
 /security-harness/blog
+
+# Harness runtime state (git worktrees + session/output dirs)
+.cc-worktrees/
+.conductor-code/

coding-harness/.python-version

@@ -0,0 +1 @@
+3.13

coding-harness/Makefile

@@ -1,30 +1,136 @@
-SHELL := /bin/bash
-PYTHON ?= python3
-VENV := .venv-ci
-PY := $(VENV)/bin/python
+# conductor coding-harness — operator shortcuts.
+# Run `make` (or `make help`) to list targets.
+
+SHELL  := /bin/bash
+ROOT   := $(shell pwd)
+
+# uv manages the Python toolchain: it provisions PY_VERSION (per .python-version /
+# pyproject requires-python), downloading it if the host lacks it, so venvs never
+# fall back to the system python3 (which has no claude-agent-sdk wheels).
+UV         := uv
+PY_VERSION := 3.13
+
+# run.sh owns the worker/tui virtualenvs (workers/.venv, tui/.venv), .env
+# loading, and server autostart; delegate to it so this Makefile never drifts.
+RUN    := $(ROOT)/run.sh
+WORKER_PY := $(ROOT)/workers/.venv/bin/python
+
+# CI virtualenv: tui + worker deps + pytest, kept separate from the operator
+# venvs so `make ci` is reproducible and never touches a running worker env.
+VENV := $(ROOT)/.venv-ci
+PY   := $(VENV)/bin/python
 DEPS := $(VENV)/.deps-installed
 
+export CONDUCTOR_SERVER_URL ?= http://localhost:8080/api
+
 .DEFAULT_GOAL := help
 
 help: ## List available targets
 	@grep -E '^[a-zA-Z_-]+:.*?## ' $(MAKEFILE_LIST) | sort | \
-	  awk 'BEGIN{FS=":.*?## "}{printf "  \033[36m%-12s\033[0m %s\n", $$1, $$2}'
+	  awk 'BEGIN{FS=":.*?## "}{printf "  \033[36m%-14s\033[0m %s\n", $$1, $$2}'
+
+# ── Conductor server ────────────────────────────────────────────────────────
+
+server: ## Start the local Conductor server (OSS; Java 21+)
+	@conductor server start
+
+server-stop: ## Stop the local Conductor server
+	@conductor server stop
+
+server-logs: ## Tail the Conductor server log
+	@conductor server logs
+
+# ── Setup ───────────────────────────────────────────────────────────────────
+
+venv: ## Create the worker virtualenv and install deps (workers/.venv)
+	@$(RUN) setup
+
+tui-venv: ## Create the TUI virtualenv and install deps (tui/.venv)
+	@test -x $(ROOT)/tui/.venv/bin/python || $(UV) venv --python $(PY_VERSION) $(ROOT)/tui/.venv
+	@$(UV) pip install -q --prerelease=explicit --python $(ROOT)/tui/.venv/bin/python -r $(ROOT)/tui/requirements.txt
+	@echo "tui venv ready: $(ROOT)/tui/.venv"
+
+# ── Definitions & workers ─────────────────────────────────────────────────────
+
+register: ## Register/update task defs + workflows (starts a local server if needed)
+	@$(RUN) register
+
+WORKER_MODULES ?= coding_agent,gitops
+workers: venv ## Run workers (supervised, foreground; WORKER_MODULES=coding_agent|gitops to split)
+	@WORKER_MODULES=$(WORKER_MODULES) $(ROOT)/workers/run_workers.sh
+
+run: ## Full bootstrap: setup + register + run workers (the one-command start)
+	@$(RUN)
+
+tui: ## Launch the terminal UI (chat; ANTHROPIC_API_KEY needed for chat)
+	@$(RUN) tui
+
+# ── Manual test: start a workflow run ─────────────────────────────────────────
+# Each target builds its --input JSON with jq (safe quoting) and hands off to
+# `conductor workflow start`. Pass extra conductor flags via ARGS='...'.
+
+code: ## Code a change in a local repo: make code REPO=/abs/path INSTRUCTION="…" [DESIGN=true]
+	@test -n "$(REPO)" && test -n "$(INSTRUCTION)" || { \
+	  echo "usage: make code REPO=/abs/path/to/repo INSTRUCTION=\"do X\" [DESIGN=true]"; exit 1; }
+	@input=$$(jq -cn --arg r "$(REPO)" --arg i "$(INSTRUCTION)" --argjson d $(if $(DESIGN),$(DESIGN),false) \
+	  '{repoPath:$$r, instruction:$$i, design:$$d}'); \
+	  conductor workflow start --workflow code_parallel --input "$$input" $(ARGS)
+
+issue-to-pr: ## GitHub issue → PR: make issue-to-pr REPO=owner/name ISSUE=7
+	@test -n "$(REPO)" && test -n "$(ISSUE)" || { \
+	  echo "usage: make issue-to-pr REPO=owner/name ISSUE=<n>"; exit 1; }
+	@input=$$(jq -cn --arg r "$(REPO)" --argjson n $(ISSUE) '{repo:$$r, issueNumber:$$n}'); \
+	  conductor workflow start --workflow issue_to_pr --input "$$input" $(ARGS)
+
+pr-review: ## Review a PR (posts a formal review): make pr-review REPO=owner/name PR=7
+	@test -n "$(REPO)" && test -n "$(PR)" || { \
+	  echo "usage: make pr-review REPO=owner/name PR=<n>"; exit 1; }
+	@input=$$(jq -cn --arg r "$(REPO)" --argjson n $(PR) '{repo:$$r, prNumber:$$n}'); \
+	  conductor workflow start --workflow pr_review --input "$$input" $(ARGS)
+
+address-pr: ## Apply a PR's review feedback: make address-pr REPO=owner/name PR=7
+	@test -n "$(REPO)" && test -n "$(PR)" || { \
+	  echo "usage: make address-pr REPO=owner/name PR=<n>"; exit 1; }
+	@input=$$(jq -cn --arg r "$(REPO)" --argjson n $(PR) '{repo:$$r, prNumber:$$n}'); \
+	  conductor workflow start --workflow address_pr --input "$$input" $(ARGS)
+
+github-demo: ## Clone → change → PR smoke test: make github-demo REPO_URL=… INSTRUCTION="…"
+	@test -n "$(REPO_URL)" && test -n "$(INSTRUCTION)" || { \
+	  echo "usage: make github-demo REPO_URL=<git-url> INSTRUCTION=\"do X\""; exit 1; }
+	@input=$$(jq -cn --arg u "$(REPO_URL)" --arg i "$(INSTRUCTION)" '{repoUrl:$$u, instruction:$$i}'); \
+	  conductor workflow start --workflow github_demo --input "$$input" $(ARGS)
+
+watch: ## Follow a running execution: make watch ID=<workflowId>
+	@test -n "$(ID)" || { echo "usage: make watch ID=<workflowId>"; exit 1; }
+	@conductor workflow get-execution "$(ID)" -c
+
+# ── Quality bar ───────────────────────────────────────────────────────────────
 
 $(DEPS):
-	@test -x $(PY) || $(PYTHON) -m venv $(VENV)
-	@$(VENV)/bin/pip install -q -r tui/requirements.txt -r workers/requirements.txt pytest pytest-asyncio
+	@test -x $(PY) || $(UV) venv --python $(PY_VERSION) $(VENV)
+	@$(UV) pip install -q --prerelease=explicit --python $(PY) -r tui/requirements.txt -r workers/requirements.txt pytest pytest-asyncio
 	@touch $(DEPS)
 
 validate: ## Validate workflow and task-definition JSON
 	@jq empty workers/workflows/*.json workers/workflows/taskdefs/*.json
 
-lint: ## Compile Python and check shell syntax
-	@$(PYTHON) -m compileall -q tui workers
+lint: $(DEPS) ## Compile Python and check shell syntax
+	@$(PY) -m compileall -q tui workers
 	@bash -n run.sh workers/register.sh workers/run_workers.sh
 
-test-ci: $(DEPS) ## Run unit tests
+test: $(DEPS) ## Run unit tests (tui + workers)
 	@$(PY) -m pytest tui/tests workers/tests -q
 
-ci: validate lint test-ci ## Run the coding-harness CI quality bar
+test-ci: test ## Alias for `test` (CI unit tests)
+
+ci: validate lint test ## Run the coding-harness CI quality bar
+
+# ── Cleanup ────────────────────────────────────────────────────────────────────
+
+clean: ## Remove virtualenvs and harness runtime state (worktrees, sessions)
+	@rm -rf $(VENV) workers/.venv tui/.venv \
+	  $(ROOT)/.cc-worktrees $(ROOT)/.conductor-code && echo "cleaned"
 
-.PHONY: help validate lint test-ci ci
+.PHONY: help server server-stop server-logs venv tui-venv register workers run tui \
+        code issue-to-pr pr-review address-pr github-demo watch \
+        validate lint test test-ci ci clean