wells-agent 0.3.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- wells_agent-0.3.0/.env.example +120 -0
- wells_agent-0.3.0/.github/workflows/binary.yml +73 -0
- wells_agent-0.3.0/.github/workflows/build-index-wheels.yml +57 -0
- wells_agent-0.3.0/.github/workflows/ci.yml +24 -0
- wells_agent-0.3.0/.github/workflows/pypi.yml +30 -0
- wells_agent-0.3.0/.github/workflows/release-index.yml +68 -0
- wells_agent-0.3.0/.gitignore +30 -0
- wells_agent-0.3.0/.gitmodules +24 -0
- wells_agent-0.3.0/.python-version +1 -0
- wells_agent-0.3.0/AGENT.md +178 -0
- wells_agent-0.3.0/AGENTS.md +81 -0
- wells_agent-0.3.0/PKG-INFO +1776 -0
- wells_agent-0.3.0/README.md +1746 -0
- wells_agent-0.3.0/RULES.md +155 -0
- wells_agent-0.3.0/about.html +401 -0
- wells_agent-0.3.0/debug_api.py +90 -0
- wells_agent-0.3.0/handoff.md +172 -0
- wells_agent-0.3.0/install-remote.ps1 +40 -0
- wells_agent-0.3.0/install-remote.sh +44 -0
- wells_agent-0.3.0/install.ps1 +24 -0
- wells_agent-0.3.0/install.sh +40 -0
- wells_agent-0.3.0/output_sample.txt +537 -0
- wells_agent-0.3.0/packaging/entry.py +10 -0
- wells_agent-0.3.0/packaging/wells.spec +72 -0
- wells_agent-0.3.0/pyproject.toml +69 -0
- wells_agent-0.3.0/scripts/evolve_handoff_to_autoloop.sh +60 -0
- wells_agent-0.3.0/scripts/wells_resumable_launch.sh +261 -0
- wells_agent-0.3.0/skills/add-provider-profile/SKILL.md +19 -0
- wells_agent-0.3.0/skills/worktree-agents/SKILL.md +84 -0
- wells_agent-0.3.0/src/wells/AGENT.md +111 -0
- wells_agent-0.3.0/src/wells/__init__.py +25 -0
- wells_agent-0.3.0/src/wells/__main__.py +10 -0
- wells_agent-0.3.0/src/wells/_gitutils.py +63 -0
- wells_agent-0.3.0/src/wells/acp_server.py +390 -0
- wells_agent-0.3.0/src/wells/agents/__init__.py +0 -0
- wells_agent-0.3.0/src/wells/agents/architect.py +58 -0
- wells_agent-0.3.0/src/wells/agents/coder.py +253 -0
- wells_agent-0.3.0/src/wells/agents/planner.py +115 -0
- wells_agent-0.3.0/src/wells/agents/reviewer.py +113 -0
- wells_agent-0.3.0/src/wells/agents/tester.py +170 -0
- wells_agent-0.3.0/src/wells/background.py +668 -0
- wells_agent-0.3.0/src/wells/browser.py +435 -0
- wells_agent-0.3.0/src/wells/builtin_skills/benchmark-before-claiming/SKILL.md +16 -0
- wells_agent-0.3.0/src/wells/builtin_skills/dependency-audit/SKILL.md +23 -0
- wells_agent-0.3.0/src/wells/builtin_skills/incremental-refactor/SKILL.md +16 -0
- wells_agent-0.3.0/src/wells/builtin_skills/project-recon/SKILL.md +18 -0
- wells_agent-0.3.0/src/wells/builtin_skills/safe-migration/SKILL.md +17 -0
- wells_agent-0.3.0/src/wells/builtin_skills/verify-external-api/SKILL.md +15 -0
- wells_agent-0.3.0/src/wells/chat.py +196 -0
- wells_agent-0.3.0/src/wells/checkers.py +223 -0
- wells_agent-0.3.0/src/wells/cli.py +2789 -0
- wells_agent-0.3.0/src/wells/codeact.py +234 -0
- wells_agent-0.3.0/src/wells/codemode.py +243 -0
- wells_agent-0.3.0/src/wells/commands.py +103 -0
- wells_agent-0.3.0/src/wells/compress.py +92 -0
- wells_agent-0.3.0/src/wells/config.py +627 -0
- wells_agent-0.3.0/src/wells/context.py +128 -0
- wells_agent-0.3.0/src/wells/control.py +291 -0
- wells_agent-0.3.0/src/wells/embeddings.py +641 -0
- wells_agent-0.3.0/src/wells/evolve/__init__.py +65 -0
- wells_agent-0.3.0/src/wells/evolve/autoloop.py +280 -0
- wells_agent-0.3.0/src/wells/evolve/corpus.py +571 -0
- wells_agent-0.3.0/src/wells/evolve/mutate.py +526 -0
- wells_agent-0.3.0/src/wells/evolve/runner.py +799 -0
- wells_agent-0.3.0/src/wells/evolve/schema.py +168 -0
- wells_agent-0.3.0/src/wells/executor.py +3008 -0
- wells_agent-0.3.0/src/wells/finisher.py +232 -0
- wells_agent-0.3.0/src/wells/fleet.py +357 -0
- wells_agent-0.3.0/src/wells/gitops.py +375 -0
- wells_agent-0.3.0/src/wells/graph.py +167 -0
- wells_agent-0.3.0/src/wells/hooks.py +215 -0
- wells_agent-0.3.0/src/wells/index_tools.py +560 -0
- wells_agent-0.3.0/src/wells/index_watcher.py +164 -0
- wells_agent-0.3.0/src/wells/logger.py +107 -0
- wells_agent-0.3.0/src/wells/logo.py +24 -0
- wells_agent-0.3.0/src/wells/main.py +1702 -0
- wells_agent-0.3.0/src/wells/mcp_client.py +448 -0
- wells_agent-0.3.0/src/wells/mcp_server.py +826 -0
- wells_agent-0.3.0/src/wells/memory.py +296 -0
- wells_agent-0.3.0/src/wells/memory_bridge.py +195 -0
- wells_agent-0.3.0/src/wells/notify.py +142 -0
- wells_agent-0.3.0/src/wells/personas.py +454 -0
- wells_agent-0.3.0/src/wells/pricing.py +109 -0
- wells_agent-0.3.0/src/wells/principles.py +117 -0
- wells_agent-0.3.0/src/wells/providers.py +462 -0
- wells_agent-0.3.0/src/wells/recovery_hints.py +121 -0
- wells_agent-0.3.0/src/wells/reflections.py +552 -0
- wells_agent-0.3.0/src/wells/repomap.py +256 -0
- wells_agent-0.3.0/src/wells/rules.py +546 -0
- wells_agent-0.3.0/src/wells/runtime.py +86 -0
- wells_agent-0.3.0/src/wells/safety.py +244 -0
- wells_agent-0.3.0/src/wells/sandbox.py +235 -0
- wells_agent-0.3.0/src/wells/schedule.py +362 -0
- wells_agent-0.3.0/src/wells/sessions.py +210 -0
- wells_agent-0.3.0/src/wells/settings.py +845 -0
- wells_agent-0.3.0/src/wells/setup.py +322 -0
- wells_agent-0.3.0/src/wells/skill_authoring.py +479 -0
- wells_agent-0.3.0/src/wells/skills.py +537 -0
- wells_agent-0.3.0/src/wells/spend_guard.py +113 -0
- wells_agent-0.3.0/src/wells/state.py +60 -0
- wells_agent-0.3.0/src/wells/subagents.py +251 -0
- wells_agent-0.3.0/src/wells/summarize.py +89 -0
- wells_agent-0.3.0/src/wells/todo.py +118 -0
- wells_agent-0.3.0/src/wells/tokens.py +277 -0
- wells_agent-0.3.0/src/wells/tools.py +1248 -0
- wells_agent-0.3.0/src/wells/traces.py +465 -0
- wells_agent-0.3.0/src/wells/tui.py +4366 -0
- wells_agent-0.3.0/src/wells/user_memory.py +303 -0
- wells_agent-0.3.0/src/wells/vision.py +164 -0
- wells_agent-0.3.0/src/wells/web.py +249 -0
- wells_agent-0.3.0/src/wells/worktree.py +245 -0
- wells_agent-0.3.0/tests/__init__.py +0 -0
- wells_agent-0.3.0/tests/conftest.py +26 -0
- wells_agent-0.3.0/tests/test_acp_server.py +371 -0
- wells_agent-0.3.0/tests/test_background.py +304 -0
- wells_agent-0.3.0/tests/test_background_worktrees.py +333 -0
- wells_agent-0.3.0/tests/test_batch_masking.py +240 -0
- wells_agent-0.3.0/tests/test_browser.py +250 -0
- wells_agent-0.3.0/tests/test_chat.py +120 -0
- wells_agent-0.3.0/tests/test_codeact.py +204 -0
- wells_agent-0.3.0/tests/test_codemode.py +119 -0
- wells_agent-0.3.0/tests/test_commands.py +135 -0
- wells_agent-0.3.0/tests/test_crossharness.py +87 -0
- wells_agent-0.3.0/tests/test_embeddings.py +403 -0
- wells_agent-0.3.0/tests/test_evolve.py +779 -0
- wells_agent-0.3.0/tests/test_evolve_autoloop.py +224 -0
- wells_agent-0.3.0/tests/test_evolve_mutate.py +323 -0
- wells_agent-0.3.0/tests/test_executor.py +1773 -0
- wells_agent-0.3.0/tests/test_fleet.py +341 -0
- wells_agent-0.3.0/tests/test_headless.py +254 -0
- wells_agent-0.3.0/tests/test_hooks.py +234 -0
- wells_agent-0.3.0/tests/test_layer3.py +259 -0
- wells_agent-0.3.0/tests/test_logger.py +39 -0
- wells_agent-0.3.0/tests/test_mcp_client.py +270 -0
- wells_agent-0.3.0/tests/test_mcp_server.py +197 -0
- wells_agent-0.3.0/tests/test_memory.py +117 -0
- wells_agent-0.3.0/tests/test_memory_bridge.py +129 -0
- wells_agent-0.3.0/tests/test_notify.py +192 -0
- wells_agent-0.3.0/tests/test_orchestrator.py +369 -0
- wells_agent-0.3.0/tests/test_personas.py +333 -0
- wells_agent-0.3.0/tests/test_platform.py +344 -0
- wells_agent-0.3.0/tests/test_principles.py +122 -0
- wells_agent-0.3.0/tests/test_providers_and_settings.py +530 -0
- wells_agent-0.3.0/tests/test_quality.py +393 -0
- wells_agent-0.3.0/tests/test_recovery_hints.py +45 -0
- wells_agent-0.3.0/tests/test_reflections.py +278 -0
- wells_agent-0.3.0/tests/test_rules.py +465 -0
- wells_agent-0.3.0/tests/test_sandbox.py +186 -0
- wells_agent-0.3.0/tests/test_schedule.py +284 -0
- wells_agent-0.3.0/tests/test_skill_authoring.py +276 -0
- wells_agent-0.3.0/tests/test_skills.py +585 -0
- wells_agent-0.3.0/tests/test_spend_guard.py +123 -0
- wells_agent-0.3.0/tests/test_stepwise.py +200 -0
- wells_agent-0.3.0/tests/test_stop_banner.py +188 -0
- wells_agent-0.3.0/tests/test_todo.py +177 -0
- wells_agent-0.3.0/tests/test_tools.py +400 -0
- wells_agent-0.3.0/tests/test_trace_analysis.py +392 -0
- wells_agent-0.3.0/tests/test_traces.py +150 -0
- wells_agent-0.3.0/tests/test_tui.py +662 -0
- wells_agent-0.3.0/tests/test_user_memory.py +237 -0
- wells_agent-0.3.0/tests/test_vision.py +129 -0
- wells_agent-0.3.0/tests/test_vision_cli.py +203 -0
- wells_agent-0.3.0/tests/test_web.py +256 -0
- wells_agent-0.3.0/uv.lock +2405 -0
- wells_agent-0.3.0/wells +109 -0
- wells_agent-0.3.0/wells-index/Cargo.lock +859 -0
- wells_agent-0.3.0/wells-index/Cargo.toml +38 -0
- wells_agent-0.3.0/wells-index/README.md +218 -0
- wells_agent-0.3.0/wells-index/build.rs +76 -0
- wells_agent-0.3.0/wells-index/pyproject.toml +33 -0
- wells_agent-0.3.0/wells-index/python/wells_index/__init__.py +12 -0
- wells_agent-0.3.0/wells-index/src/detect.rs +84 -0
- wells_agent-0.3.0/wells-index/src/engine.rs +147 -0
- wells_agent-0.3.0/wells-index/src/graph.rs +127 -0
- wells_agent-0.3.0/wells-index/src/hash.rs +17 -0
- wells_agent-0.3.0/wells-index/src/lib.rs +175 -0
- wells_agent-0.3.0/wells-index/src/parse.rs +535 -0
- wells_agent-0.3.0/wells-index/src/query.rs +20 -0
- wells_agent-0.3.0/wells-index/src/store.rs +377 -0
- wells_agent-0.3.0/wells.bat +117 -0
- wells_agent-0.3.0/wells_logo.png +0 -0
|
@@ -0,0 +1,120 @@
|
|
|
1
|
+
# Wells harness — example configuration.
|
|
2
|
+
# Copy to .env and edit. Or run `wells config` for an interactive menu.
|
|
3
|
+
|
|
4
|
+
# ===========================================================================
|
|
5
|
+
# Provider profiles (model-agnostic core)
|
|
6
|
+
# ===========================================================================
|
|
7
|
+
# A *profile* is a named bundle of (provider kind, model, api key, base url).
|
|
8
|
+
# Configure each profile with MODEL_<name> / API_KEY_<name> / BASE_URL_<name>.
|
|
9
|
+
|
|
10
|
+
# Which profiles are configured (comma-separated) and which is active.
|
|
11
|
+
MODEL_PROFILES=zai
|
|
12
|
+
MODEL_PROFILE=zai
|
|
13
|
+
# Optional: cheaper model for low-stakes subtasks (summarize/compress/etc).
|
|
14
|
+
# MODEL_PROFILE_CHEAP=zai
|
|
15
|
+
# Optional: vision-capable profile, routed to automatically when a task has
|
|
16
|
+
# image attachments and the active profile isn't vision-capable (see
|
|
17
|
+
# wells.vision.provider_supports_vision). Blank = same as active. Manage
|
|
18
|
+
# from `wells config` -> "v) Add / switch / clear vision provider profile".
|
|
19
|
+
# MODEL_PROFILE_VISION=openrouter
|
|
20
|
+
|
|
21
|
+
# --- Built-in `zai` profile (Z.ai GLM, OpenAI-compatible) ------------------
|
|
22
|
+
# Uses the CODING endpoint (/api/coding/paas/v4/) required by GLM-4.x/5.x.
|
|
23
|
+
# NOTE: legacy ZAI_* vars below ALSO work and seed this same profile; explicit
|
|
24
|
+
# MODEL_zai / BASE_URL_zai / API_KEY_zai take precedence when both are set.
|
|
25
|
+
MODEL_zai=glm-5.2
|
|
26
|
+
API_KEY_zai=your-zai-api-key-here
|
|
27
|
+
BASE_URL_zai=https://api.z.ai/api/coding/paas/v4/
|
|
28
|
+
|
|
29
|
+
# --- Example: add more profiles (uncomment + edit) -------------------------
|
|
30
|
+
# Google Gemini (requires: pip install langchain-google-genai):
|
|
31
|
+
# MODEL_gemini=gemini-2.0-flash
|
|
32
|
+
# API_KEY_gemini=your-gemini-api-key
|
|
33
|
+
#
|
|
34
|
+
# OpenAI direct:
|
|
35
|
+
# MODEL_OPENAI=gpt-5o-mini
|
|
36
|
+
# API_KEY_OPENAI=sk-...
|
|
37
|
+
#
|
|
38
|
+
# OpenRouter (hundreds of models via one key; also reads OPENROUTER_API_KEY
|
|
39
|
+
# automatically, so you don't need to duplicate it as API_KEY_openrouter):
|
|
40
|
+
# MODEL_openrouter=anthropic/claude-sonnet-4-5
|
|
41
|
+
# API_KEY_openrouter=sk-or-...
|
|
42
|
+
#
|
|
43
|
+
# OpenRouter free vision model, as a dedicated vision profile (:free models
|
|
44
|
+
# have low rate limits and vary in availability -- check openrouter.ai/models
|
|
45
|
+
# for the current :free vision lineup):
|
|
46
|
+
# MODEL_openrouter=nvidia/nemotron-nano-12b-v2-vl:free
|
|
47
|
+
# MODEL_PROFILE_VISION=openrouter
|
|
48
|
+
#
|
|
49
|
+
# Anthropic direct (requires: pip install langchain-anthropic):
|
|
50
|
+
# MODEL_anthropic=claude-sonnet-4-5
|
|
51
|
+
# API_KEY_anthropic=sk-ant-...
|
|
52
|
+
#
|
|
53
|
+
# Local Ollama (requires: pip install langchain-ollama):
|
|
54
|
+
# MODEL_ollama=devstral
|
|
55
|
+
# BASE_URL_ollama=http://localhost:11434
|
|
56
|
+
#
|
|
57
|
+
# Any OpenAI-compatible local endpoint (vLLM, LM Studio, etc.):
|
|
58
|
+
# MODEL_local=my-finetune
|
|
59
|
+
# BASE_URL_local=http://localhost:8000/v1
|
|
60
|
+
|
|
61
|
+
# ===========================================================================
|
|
62
|
+
# Legacy Z.ai variables (still supported — seeds the `zai` profile)
|
|
63
|
+
# ===========================================================================
|
|
64
|
+
# ZAI_API_KEY=your-zai-api-key-here
|
|
65
|
+
# ZAI_ENDPOINT=https://api.z.ai/api/coding/paas/v4/
|
|
66
|
+
# ZAI_MODEL=glm-5.2
|
|
67
|
+
# ZAI_MODEL_CHEAP=
|
|
68
|
+
|
|
69
|
+
# ===========================================================================
|
|
70
|
+
# Workspace + safety
|
|
71
|
+
# ===========================================================================
|
|
72
|
+
# Directory the agent is confined to (path escapes are blocked).
|
|
73
|
+
# Defaults to the current working directory.
|
|
74
|
+
# WORKSPACE_ROOT=/path/to/your/project
|
|
75
|
+
|
|
76
|
+
# Safety policy for writes/shell: auto | approve | dryrun
|
|
77
|
+
# auto - execute immediately (confined to WORKSPACE_ROOT)
|
|
78
|
+
# approve - require an approval callback (dry-run if none wired)
|
|
79
|
+
# dryrun - never execute; describe what would happen
|
|
80
|
+
HARNESS_SAFETY=auto
|
|
81
|
+
|
|
82
|
+
# Plan mode: coder plans edits without applying them (review-first workflow).
|
|
83
|
+
PLAN_MODE=0
|
|
84
|
+
|
|
85
|
+
# ===========================================================================
|
|
86
|
+
# Run behaviour
|
|
87
|
+
# ===========================================================================
|
|
88
|
+
# Max coder <-> tester <-> reviewer loop iterations.
|
|
89
|
+
MAX_ITERATIONS=3
|
|
90
|
+
# Max tool-call rounds in a single agentic executor run.
|
|
91
|
+
MAX_TOOL_STEPS=25
|
|
92
|
+
# Max seconds for a single shell command.
|
|
93
|
+
SHELL_TIMEOUT=120
|
|
94
|
+
# When 1, the finisher pushes the branch and opens a PR via `gh` after a run.
|
|
95
|
+
WELLS_OPEN_PR=0
|
|
96
|
+
|
|
97
|
+
# ===========================================================================
|
|
98
|
+
# Token optimization
|
|
99
|
+
# ===========================================================================
|
|
100
|
+
TOKEN_BUDGET_MAX_INPUT=24000
|
|
101
|
+
TOKEN_BUDGET_SMALL_INPUT=8000
|
|
102
|
+
TOKEN_BUDGET_RESERVED_OUTPUT=4000
|
|
103
|
+
# Replace verbatim plan/architecture with a summary on loop iterations once the
|
|
104
|
+
# durable context exceeds this many tokens. Set 0 or SUMMARIZE_ON_LOOP=false
|
|
105
|
+
# to disable (keeps full context verbatim every iteration).
|
|
106
|
+
SUMMARIZE_ON_LOOP=1
|
|
107
|
+
SUMMARIZE_THRESHOLD=1500
|
|
108
|
+
|
|
109
|
+
# ===========================================================================
|
|
110
|
+
# LLM tuning
|
|
111
|
+
# ===========================================================================
|
|
112
|
+
LLM_TIMEOUT=180
|
|
113
|
+
LLM_MAX_RETRIES=5
|
|
114
|
+
LLM_BACKOFF_BASE=2.0
|
|
115
|
+
|
|
116
|
+
# ===========================================================================
|
|
117
|
+
# Command screening (optional)
|
|
118
|
+
# ===========================================================================
|
|
119
|
+
# |-separated regex patterns always refused by run_command, regardless of safety.
|
|
120
|
+
# BLOCKED_COMMANDS=rm\s+-rf\s+/|mkfs|dd\s+if=|:\(\)\s*\{|shutdown|reboot
|
|
@@ -0,0 +1,73 @@
|
|
|
1
|
+
name: Build standalone binaries
|
|
2
|
+
|
|
3
|
+
# Produces the "no Python required" release artifacts: a onedir PyInstaller
|
|
4
|
+
# bundle per platform, zipped and attached to the GitHub Release for the tag.
|
|
5
|
+
# These are what the installers fall back to when a user wants an instant
|
|
6
|
+
# cold start instead of a uv-managed toolchain.
|
|
7
|
+
#
|
|
8
|
+
# push tag v* -> PyPI publish (pypi.yml) + binaries (this file)
|
|
9
|
+
|
|
10
|
+
on:
|
|
11
|
+
push:
|
|
12
|
+
tags: ["v*"]
|
|
13
|
+
|
|
14
|
+
permissions:
|
|
15
|
+
contents: write
|
|
16
|
+
|
|
17
|
+
jobs:
|
|
18
|
+
build:
|
|
19
|
+
name: ${{ matrix.label }}
|
|
20
|
+
runs-on: ${{ matrix.os }}
|
|
21
|
+
strategy:
|
|
22
|
+
fail-fast: false
|
|
23
|
+
matrix:
|
|
24
|
+
include:
|
|
25
|
+
- os: windows-latest
|
|
26
|
+
label: windows-x64
|
|
27
|
+
- os: ubuntu-latest
|
|
28
|
+
label: linux-x64
|
|
29
|
+
- os: macos-14
|
|
30
|
+
label: macos-arm64
|
|
31
|
+
- os: macos-13
|
|
32
|
+
label: macos-x64
|
|
33
|
+
|
|
34
|
+
steps:
|
|
35
|
+
- uses: actions/checkout@v4
|
|
36
|
+
|
|
37
|
+
- uses: actions/setup-python@v5
|
|
38
|
+
with:
|
|
39
|
+
python-version: "3.12"
|
|
40
|
+
|
|
41
|
+
- name: Install the project + PyInstaller
|
|
42
|
+
run: |
|
|
43
|
+
python -m pip install --upgrade pip
|
|
44
|
+
python -m pip install -e .
|
|
45
|
+
python -m pip install pyinstaller
|
|
46
|
+
|
|
47
|
+
- name: Build onedir bundle
|
|
48
|
+
run: pyinstaller packaging/wells.spec --noconfirm --clean
|
|
49
|
+
|
|
50
|
+
- name: Smoke test the binary
|
|
51
|
+
shell: bash
|
|
52
|
+
run: |
|
|
53
|
+
./dist/wells/wells --version || ./dist/wells/wells.exe --version
|
|
54
|
+
|
|
55
|
+
- name: Package
|
|
56
|
+
shell: bash
|
|
57
|
+
run: |
|
|
58
|
+
VER="${GITHUB_REF_NAME#v}"
|
|
59
|
+
OUT="wells-${VER}-${{ matrix.label }}"
|
|
60
|
+
if [ "${{ runner.os }}" = "Windows" ]; then
|
|
61
|
+
(cd dist && 7z a -tzip "../${OUT}.zip" wells > /dev/null)
|
|
62
|
+
else
|
|
63
|
+
(cd dist && tar -czf "../${OUT}.tar.gz" wells)
|
|
64
|
+
fi
|
|
65
|
+
ls -la "${OUT}".* || true
|
|
66
|
+
|
|
67
|
+
- name: Attach to release
|
|
68
|
+
uses: softprops/action-gh-release@v2
|
|
69
|
+
with:
|
|
70
|
+
files: |
|
|
71
|
+
wells-*.zip
|
|
72
|
+
wells-*.tar.gz
|
|
73
|
+
fail_on_unmatched_files: false
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
name: Build wells-index wheels
|
|
2
|
+
|
|
3
|
+
on:
|
|
4
|
+
push:
|
|
5
|
+
tags:
|
|
6
|
+
- "wells-index-v*"
|
|
7
|
+
|
|
8
|
+
jobs:
|
|
9
|
+
build:
|
|
10
|
+
runs-on: ${{ matrix.os }}
|
|
11
|
+
strategy:
|
|
12
|
+
matrix:
|
|
13
|
+
os: [ubuntu-latest, macos-latest, windows-latest]
|
|
14
|
+
python-version: ["3.12"]
|
|
15
|
+
defaults:
|
|
16
|
+
run:
|
|
17
|
+
working-directory: wells-index
|
|
18
|
+
|
|
19
|
+
steps:
|
|
20
|
+
- uses: actions/checkout@v4
|
|
21
|
+
with:
|
|
22
|
+
submodules: recursive
|
|
23
|
+
|
|
24
|
+
- uses: actions/setup-python@v5
|
|
25
|
+
with:
|
|
26
|
+
python-version: ${{ matrix.python-version }}
|
|
27
|
+
|
|
28
|
+
- name: Install Rust
|
|
29
|
+
uses: dtolnay/rust-toolchain@stable
|
|
30
|
+
|
|
31
|
+
- name: Install maturin
|
|
32
|
+
run: pip install maturin
|
|
33
|
+
|
|
34
|
+
- name: Build wheels
|
|
35
|
+
run: maturin build --release
|
|
36
|
+
|
|
37
|
+
- name: Upload artifacts
|
|
38
|
+
uses: actions/upload-artifact@v4
|
|
39
|
+
with:
|
|
40
|
+
name: wheels-${{ matrix.os }}
|
|
41
|
+
path: wells-index/target/wheels/
|
|
42
|
+
|
|
43
|
+
publish:
|
|
44
|
+
needs: build
|
|
45
|
+
runs-on: ubuntu-latest
|
|
46
|
+
if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')
|
|
47
|
+
|
|
48
|
+
steps:
|
|
49
|
+
- uses: actions/download-artifact@v4
|
|
50
|
+
with:
|
|
51
|
+
merge-multiple: true
|
|
52
|
+
path: wheels/
|
|
53
|
+
|
|
54
|
+
- uses: pypa/gh-action-pypi-publish@release/v1
|
|
55
|
+
with:
|
|
56
|
+
password: ${{ secrets.PYPI_API_TOKEN }}
|
|
57
|
+
packages-dir: wheels/
|
|
@@ -0,0 +1,24 @@
|
|
|
1
|
+
name: CI
|
|
2
|
+
|
|
3
|
+
on:
|
|
4
|
+
push:
|
|
5
|
+
branches: [master]
|
|
6
|
+
pull_request:
|
|
7
|
+
branches: [master]
|
|
8
|
+
|
|
9
|
+
jobs:
|
|
10
|
+
test:
|
|
11
|
+
runs-on: ubuntu-latest
|
|
12
|
+
steps:
|
|
13
|
+
- uses: actions/checkout@v4
|
|
14
|
+
|
|
15
|
+
- name: Install uv
|
|
16
|
+
uses: astral-sh/setup-uv@v5
|
|
17
|
+
with:
|
|
18
|
+
python-version: "3.12"
|
|
19
|
+
|
|
20
|
+
- name: Sync dependencies
|
|
21
|
+
run: uv sync
|
|
22
|
+
|
|
23
|
+
- name: Run tests
|
|
24
|
+
run: uv run pytest -q
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
name: Publish to PyPI
|
|
2
|
+
|
|
3
|
+
# Trusted publishing (OIDC): no API tokens. Configure once at
|
|
4
|
+
# https://pypi.org/manage/account/publishing/ :
|
|
5
|
+
# PyPI project name: wells-agent
|
|
6
|
+
# owner: corbybender repo: Wells workflow: pypi.yml environment: pypi
|
|
7
|
+
# Then pushing a v* tag builds and publishes.
|
|
8
|
+
|
|
9
|
+
on:
|
|
10
|
+
push:
|
|
11
|
+
tags: ["v*"]
|
|
12
|
+
|
|
13
|
+
permissions:
|
|
14
|
+
id-token: write
|
|
15
|
+
|
|
16
|
+
jobs:
|
|
17
|
+
build-and-publish:
|
|
18
|
+
runs-on: ubuntu-latest
|
|
19
|
+
environment: pypi
|
|
20
|
+
steps:
|
|
21
|
+
- uses: actions/checkout@v4
|
|
22
|
+
- uses: astral-sh/setup-uv@v5
|
|
23
|
+
with:
|
|
24
|
+
python-version: "3.12"
|
|
25
|
+
- name: Build sdist + wheel
|
|
26
|
+
run: uv build
|
|
27
|
+
- name: Check metadata
|
|
28
|
+
run: uvx twine check dist/*
|
|
29
|
+
- name: Publish to PyPI
|
|
30
|
+
uses: pypa/gh-action-pypi-publish@release/v1
|
|
@@ -0,0 +1,68 @@
|
|
|
1
|
+
name: Release wells-index
|
|
2
|
+
|
|
3
|
+
# Builds wells-index wheels for Linux/macOS/Windows on Python 3.12 + 3.13 and
|
|
4
|
+
# publishes them to PyPI. Trigger by pushing a tag like `index-v0.1.1`, or run
|
|
5
|
+
# manually from the Actions tab (workflow_dispatch builds without publishing
|
|
6
|
+
# unless "publish" is set to true).
|
|
7
|
+
#
|
|
8
|
+
# Publishing auth: PyPI "trusted publishing" (configure this repo + workflow
|
|
9
|
+
# as a trusted publisher for the wells-index project on pypi.org — no token
|
|
10
|
+
# needed). The pypi environment + id-token permission below are required.
|
|
11
|
+
|
|
12
|
+
on:
|
|
13
|
+
push:
|
|
14
|
+
tags: ["index-v*"]
|
|
15
|
+
workflow_dispatch:
|
|
16
|
+
inputs:
|
|
17
|
+
publish:
|
|
18
|
+
description: "Publish to PyPI"
|
|
19
|
+
type: boolean
|
|
20
|
+
default: false
|
|
21
|
+
|
|
22
|
+
jobs:
|
|
23
|
+
build:
|
|
24
|
+
strategy:
|
|
25
|
+
matrix:
|
|
26
|
+
os: [ubuntu-latest, macos-latest, windows-latest]
|
|
27
|
+
runs-on: ${{ matrix.os }}
|
|
28
|
+
steps:
|
|
29
|
+
- uses: actions/checkout@v4
|
|
30
|
+
|
|
31
|
+
- uses: actions/setup-python@v5
|
|
32
|
+
with:
|
|
33
|
+
python-version: |
|
|
34
|
+
3.12
|
|
35
|
+
3.13
|
|
36
|
+
|
|
37
|
+
- name: Build wheels
|
|
38
|
+
uses: PyO3/maturin-action@v1
|
|
39
|
+
with:
|
|
40
|
+
working-directory: wells-index
|
|
41
|
+
command: build
|
|
42
|
+
args: --release --interpreter python3.12 python3.13
|
|
43
|
+
manylinux: auto
|
|
44
|
+
|
|
45
|
+
- uses: actions/upload-artifact@v4
|
|
46
|
+
with:
|
|
47
|
+
name: wheels-${{ matrix.os }}
|
|
48
|
+
path: wells-index/target/wheels/*.whl
|
|
49
|
+
|
|
50
|
+
publish:
|
|
51
|
+
needs: build
|
|
52
|
+
if: startsWith(github.ref, 'refs/tags/index-v') || inputs.publish
|
|
53
|
+
runs-on: ubuntu-latest
|
|
54
|
+
environment: pypi
|
|
55
|
+
permissions:
|
|
56
|
+
id-token: write
|
|
57
|
+
steps:
|
|
58
|
+
- uses: actions/download-artifact@v4
|
|
59
|
+
with:
|
|
60
|
+
pattern: wheels-*
|
|
61
|
+
merge-multiple: true
|
|
62
|
+
path: dist
|
|
63
|
+
|
|
64
|
+
- name: Publish to PyPI
|
|
65
|
+
uses: pypa/gh-action-pypi-publish@release/v1
|
|
66
|
+
with:
|
|
67
|
+
packages-dir: dist
|
|
68
|
+
skip-existing: true
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
# Python-generated files
|
|
2
|
+
__pycache__/
|
|
3
|
+
*.py[oc]
|
|
4
|
+
build/
|
|
5
|
+
dist/
|
|
6
|
+
wheels/
|
|
7
|
+
*.egg-info
|
|
8
|
+
|
|
9
|
+
# Virtual environments
|
|
10
|
+
.venv
|
|
11
|
+
|
|
12
|
+
# Environment files
|
|
13
|
+
.env
|
|
14
|
+
|
|
15
|
+
# Local model caches
|
|
16
|
+
.fastembed_cache/
|
|
17
|
+
|
|
18
|
+
# Generated repo index + Rust build artifacts
|
|
19
|
+
.wells_index/
|
|
20
|
+
wells-index/target/
|
|
21
|
+
|
|
22
|
+
# Windows reserved-name artifact
|
|
23
|
+
nul
|
|
24
|
+
|
|
25
|
+
# Runtime traces
|
|
26
|
+
.wells/
|
|
27
|
+
|
|
28
|
+
# Watchdog/launch logs (scripts/wells_resumable_launch.sh) — force-add
|
|
29
|
+
# specific ones deliberately kept as evidence, same convention as .wells/
|
|
30
|
+
logs/
|
|
@@ -0,0 +1,24 @@
|
|
|
1
|
+
[submodule "wells-index/grammars/python"]
|
|
2
|
+
path = wells-index/grammars/python
|
|
3
|
+
url = https://github.com/tree-sitter/tree-sitter-python.git
|
|
4
|
+
[submodule "wells-index/grammars/javascript"]
|
|
5
|
+
path = wells-index/grammars/javascript
|
|
6
|
+
url = https://github.com/tree-sitter/tree-sitter-javascript.git
|
|
7
|
+
[submodule "wells-index/grammars/typescript"]
|
|
8
|
+
path = wells-index/grammars/typescript
|
|
9
|
+
url = https://github.com/tree-sitter/tree-sitter-typescript.git
|
|
10
|
+
[submodule "wells-index/grammars/go"]
|
|
11
|
+
path = wells-index/grammars/go
|
|
12
|
+
url = https://github.com/tree-sitter/tree-sitter-go.git
|
|
13
|
+
[submodule "wells-index/grammars/rust"]
|
|
14
|
+
path = wells-index/grammars/rust
|
|
15
|
+
url = https://github.com/tree-sitter/tree-sitter-rust.git
|
|
16
|
+
[submodule "wells-index/grammars/java"]
|
|
17
|
+
path = wells-index/grammars/java
|
|
18
|
+
url = https://github.com/tree-sitter/tree-sitter-java.git
|
|
19
|
+
[submodule "wells-index/grammars/c"]
|
|
20
|
+
path = wells-index/grammars/c
|
|
21
|
+
url = https://github.com/tree-sitter/tree-sitter-c.git
|
|
22
|
+
[submodule "wells-index/grammars/cpp"]
|
|
23
|
+
path = wells-index/grammars/cpp
|
|
24
|
+
url = https://github.com/tree-sitter/tree-sitter-cpp.git
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
3.12
|
|
@@ -0,0 +1,178 @@
|
|
|
1
|
+
# AGENT.md
|
|
2
|
+
|
|
3
|
+
> A practical set of operating principles for AI coding agents, inspired
|
|
4
|
+
> by the public four-rule CLAUDE.md philosophy and expanded with modern
|
|
5
|
+
> agent engineering practices. This is **not** an official Karpathy
|
|
6
|
+
> document, but a community-informed synthesis.
|
|
7
|
+
|
|
8
|
+
## 1. Think Before Coding
|
|
9
|
+
|
|
10
|
+
- State assumptions before implementation.
|
|
11
|
+
- Surface tradeoffs and constraints.
|
|
12
|
+
- Ask clarifying questions instead of guessing.
|
|
13
|
+
- If no interactive user is available (bench runs, CI, autonomous
|
|
14
|
+
tasks), do not stall on the question — record the assumption you are
|
|
15
|
+
making, proceed with the safest interpretation, and state it in the
|
|
16
|
+
final report.
|
|
17
|
+
- Recommend a simpler approach when appropriate.
|
|
18
|
+
|
|
19
|
+
## 2. Simplicity First
|
|
20
|
+
|
|
21
|
+
- Write the minimum code necessary.
|
|
22
|
+
- Avoid speculative abstractions.
|
|
23
|
+
- Do not future-proof unless requested.
|
|
24
|
+
- Prefer readability over cleverness.
|
|
25
|
+
|
|
26
|
+
## 3. Surgical Changes
|
|
27
|
+
|
|
28
|
+
- Change only what the task requires.
|
|
29
|
+
- Avoid unrelated refactors or cleanup.
|
|
30
|
+
- Match the project's existing style and architecture.
|
|
31
|
+
|
|
32
|
+
## 4. Goal-Driven Execution
|
|
33
|
+
|
|
34
|
+
- Define success before writing code.
|
|
35
|
+
- Verify that the requested outcome is achieved.
|
|
36
|
+
- Stop when the goal is complete.
|
|
37
|
+
- Name the concrete deliverable before starting work — usually one
|
|
38
|
+
edit in one file. Everything else around it (new tests, changelog
|
|
39
|
+
entries, docs) is secondary, and comes only after that edit is on
|
|
40
|
+
disk.
|
|
41
|
+
|
|
42
|
+
## 5. Deterministic First
|
|
43
|
+
|
|
44
|
+
Use traditional code whenever logic can be deterministic.
|
|
45
|
+
|
|
46
|
+
LLMs are best for: - Drafting - Summarization - Classification -
|
|
47
|
+
Extraction - Reasoning over unstructured information
|
|
48
|
+
|
|
49
|
+
Use deterministic code for: - Business rules - Routing - Validation -
|
|
50
|
+
Retries - Persistence - Authorization
|
|
51
|
+
|
|
52
|
+
## 6. Budget Everything
|
|
53
|
+
|
|
54
|
+
Every agent has limits. - Maximum tokens - Maximum cost - Maximum
|
|
55
|
+
runtime - Maximum retries
|
|
56
|
+
|
|
57
|
+
Fail explicitly instead of running indefinitely — and fail early:
|
|
58
|
+
|
|
59
|
+
- Treat the wall-clock deadline as a first-class constraint. Before
|
|
60
|
+
starting any long step (a big read, a full test suite, a retry
|
|
61
|
+
loop), estimate whether it can plausibly finish in the remaining
|
|
62
|
+
time; if it cannot, split it or skip to a cheaper check.
|
|
63
|
+
- Checkpoint progress. Write the best-so-far result — including what
|
|
64
|
+
is verified and what is not — before the clock runs out. A
|
|
65
|
+
delivered, honestly-labeled partial result beats a timeout with
|
|
66
|
+
nothing.
|
|
67
|
+
- Never let a single command, verification pass, or investigation loop
|
|
68
|
+
consume the whole budget.
|
|
69
|
+
- Budget by phase, roughly: no more than a third of the run on
|
|
70
|
+
investigation, half on making the change, and keep the final fifth
|
|
71
|
+
in reserve for verification and writing up. If investigation is
|
|
72
|
+
still going past its share, stop reading and act on what is known.
|
|
73
|
+
- Investigate cheaply: prefer indexed symbol lookups over grepping
|
|
74
|
+
whole files, read only the lines you need, never re-read a file you
|
|
75
|
+
have already read, and batch independent lookups into one round
|
|
76
|
+
trip.
|
|
77
|
+
- Land the edit before perfecting it. A concrete change already
|
|
78
|
+
written to disk beats a flawless plan still being polished when the
|
|
79
|
+
clock runs out. Draft the change as soon as you know the file and
|
|
80
|
+
the lines, then refine if time remains.
|
|
81
|
+
- Give every command a hard timeout, and never re-run a command that
|
|
82
|
+
already succeeded just to see it again.
|
|
83
|
+
- Match effort to the size of the ask. A patch of a few lines should
|
|
84
|
+
take minutes end to end; when peripheral work — environment setup,
|
|
85
|
+
installing dependencies, authoring new tests, updating changelogs —
|
|
86
|
+
grows bigger than the change itself, cut the peripheral work, not
|
|
87
|
+
the change.
|
|
88
|
+
- Treat a setup or test command that fails twice as blocked: stop
|
|
89
|
+
retrying it, make the edit anyway, and report which verification
|
|
90
|
+
could not run. Never spend the delivery budget unblocking a checker.
|
|
91
|
+
|
|
92
|
+
## 7. Verify Before Trust
|
|
93
|
+
|
|
94
|
+
Treat every model output as a hypothesis.
|
|
95
|
+
|
|
96
|
+
Before making impactful changes: - Run tests - Validate outputs -
|
|
97
|
+
Confirm assumptions - Prefer automated verification
|
|
98
|
+
|
|
99
|
+
Verification is budgeted, not repeated: run the exact failing check (and
|
|
100
|
+
one rerun per fix iteration) rather than re-running broad suites on a
|
|
101
|
+
loop.
|
|
102
|
+
|
|
103
|
+
Run the cheapest check that can actually fail. One targeted test, a
|
|
104
|
+
snippet against the real API, or a compile of the touched file usually
|
|
105
|
+
settles the question; a full suite or a hand-built reproduction harness
|
|
106
|
+
that costs more than the fix itself is scope creep, not verification.
|
|
107
|
+
If building the reproduction would eat the implementation budget, make
|
|
108
|
+
the change, verify what you cheaply can, and say what went unverified.
|
|
109
|
+
|
|
110
|
+
## 8. Fail Loud
|
|
111
|
+
|
|
112
|
+
Never silently continue after uncertainty.
|
|
113
|
+
|
|
114
|
+
If confidence is low: - Explain why - Present options - Ask for
|
|
115
|
+
clarification
|
|
116
|
+
|
|
117
|
+
Avoid confident but incorrect behavior.
|
|
118
|
+
|
|
119
|
+
## 9. Isolate Side Effects
|
|
120
|
+
|
|
121
|
+
Separate reasoning from execution.
|
|
122
|
+
|
|
123
|
+
The harness gates high-risk side effects via its safety system (the
|
|
124
|
+
`HARNESS_SAFETY` setting). When the harness permits an action — including
|
|
125
|
+
deployments, publishing, and external API calls — the agent **must execute
|
|
126
|
+
it** using the available tools. Do not refuse or simulate an action that the
|
|
127
|
+
harness has allowed; that decision already happened upstream.
|
|
128
|
+
|
|
129
|
+
The harness is responsible for: - Blocking destructive commands - Requiring
|
|
130
|
+
approval for sensitive operations - Constraining the workspace
|
|
131
|
+
|
|
132
|
+
The agent is responsible for: - Reading the right files before acting -
|
|
133
|
+
Running commands and observing the actual output - Reporting results
|
|
134
|
+
accurately, including failures
|
|
135
|
+
|
|
136
|
+
## 10. Check Before Declaring Done
|
|
137
|
+
|
|
138
|
+
Before finishing, confirm: - The request was fully addressed. - No
|
|
139
|
+
unnecessary complexity was introduced. - Only relevant code was
|
|
140
|
+
changed. - The solution was verified. - Remaining assumptions are
|
|
141
|
+
documented. - The result was delivered inside the time budget — if it
|
|
142
|
+
wasn't, say so explicitly rather than letting the run time out silently.
|
|
143
|
+
|
|
144
|
+
If the task started from a specific failing check — a failing test, a
|
|
145
|
+
reproducible bug, an error message — re-run that *exact* check one more
|
|
146
|
+
time immediately before declaring done. A fix that hasn't been
|
|
147
|
+
confirmed against the specific thing it was meant to fix is not
|
|
148
|
+
verified, no matter how confident the reasoning behind it looks.
|
|
149
|
+
Passing a *different* or *related* check is not the same as passing
|
|
150
|
+
*this* one.
|
|
151
|
+
|
|
152
|
+
Delivering the change outranks verifying it. When the clock is nearly
|
|
153
|
+
out, write the edit first and report plainly which checks ran and which
|
|
154
|
+
did not — an unverified diff in hand is recoverable, a timeout with
|
|
155
|
+
nothing on disk is not.
|
|
156
|
+
|
|
157
|
+
## 11. Evidence Over Confidence
|
|
158
|
+
|
|
159
|
+
Always distinguish between:
|
|
160
|
+
|
|
161
|
+
- **Observed** --- verified directly.
|
|
162
|
+
- **Inferred** --- logically concluded.
|
|
163
|
+
- **Hypothesized** --- plausible but unverified.
|
|
164
|
+
- **Recommended** --- suggested next action.
|
|
165
|
+
|
|
166
|
+
Never claim to have: - Run tests you did not run. - Read files you did
|
|
167
|
+
not inspect. - Verified behavior you did not verify. - Reproduced bugs
|
|
168
|
+
you did not reproduce.
|
|
169
|
+
|
|
170
|
+
Trust is built through evidence, not confidence.
|
|
171
|
+
|
|
172
|
+
------------------------------------------------------------------------
|
|
173
|
+
|
|
174
|
+
**Guiding Principle**
|
|
175
|
+
|
|
176
|
+
> Slow down. Think clearly. Change as little as necessary. Verify
|
|
177
|
+
> everything. Be explicit about uncertainty. Land the change early, and
|
|
178
|
+
> finish inside the budget.
|