okf-generator 0.1.1__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.
- okf_generator-0.1.1/LICENSE +9 -0
- okf_generator-0.1.1/PKG-INFO +320 -0
- okf_generator-0.1.1/README.md +273 -0
- okf_generator-0.1.1/SKILL.md +220 -0
- okf_generator-0.1.1/okf/__init__.py +34 -0
- okf_generator-0.1.1/okf/cli.py +58 -0
- okf_generator-0.1.1/okf/generator.py +1626 -0
- okf_generator-0.1.1/okf/lookup.py +396 -0
- okf_generator-0.1.1/okf/pairs.py +746 -0
- okf_generator-0.1.1/pyproject.toml +83 -0
- okf_generator-0.1.1/references/opencode-integration.md +102 -0
- okf_generator-0.1.1/tests/__init__.py +0 -0
- okf_generator-0.1.1/tests/fixtures/sample_codebase/connector.py +12 -0
- okf_generator-0.1.1/tests/fixtures/sample_codebase/utils.py +25 -0
- okf_generator-0.1.1/tests/test_generator.py +161 -0
- okf_generator-0.1.1/tests/test_lookup.py +159 -0
|
@@ -0,0 +1,9 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 Umair Baig
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
|
|
6
|
+
|
|
7
|
+
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
|
|
8
|
+
|
|
9
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
|
@@ -0,0 +1,320 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: okf-generator
|
|
3
|
+
Version: 0.1.1
|
|
4
|
+
Summary: Generate OKF v0.1 knowledge bundles from codebases — Claude skill + OpenCode integration
|
|
5
|
+
Project-URL: Homepage, https://github.com/umairbaig/okf-generator
|
|
6
|
+
Project-URL: Documentation, https://github.com/umairbaig/okf-generator#readme
|
|
7
|
+
Project-URL: Repository, https://github.com/umairbaig/okf-generator
|
|
8
|
+
Project-URL: Issues, https://github.com/umairbaig/okf-generator/issues
|
|
9
|
+
Author: Umair Baig
|
|
10
|
+
License: MIT
|
|
11
|
+
License-File: LICENSE
|
|
12
|
+
Keywords: ai-agent,claude-skill,code-knowledge,codebase-indexing,llm,okf,open-knowledge-format,opencode,training-data,tree-sitter
|
|
13
|
+
Classifier: Development Status :: 3 - Alpha
|
|
14
|
+
Classifier: Intended Audience :: Developers
|
|
15
|
+
Classifier: License :: OSI Approved :: MIT License
|
|
16
|
+
Classifier: Programming Language :: Python :: 3
|
|
17
|
+
Classifier: Programming Language :: Python :: 3.11
|
|
18
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
19
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
20
|
+
Classifier: Topic :: Software Development :: Documentation
|
|
21
|
+
Classifier: Topic :: Software Development :: Libraries :: Python Modules
|
|
22
|
+
Requires-Python: >=3.11
|
|
23
|
+
Requires-Dist: pyyaml>=6.0
|
|
24
|
+
Requires-Dist: tqdm>=4.0
|
|
25
|
+
Requires-Dist: tree-sitter-go>=0.23.0
|
|
26
|
+
Requires-Dist: tree-sitter-java>=0.23.0
|
|
27
|
+
Requires-Dist: tree-sitter-javascript>=0.23.0
|
|
28
|
+
Requires-Dist: tree-sitter-python>=0.25.0
|
|
29
|
+
Requires-Dist: tree-sitter-ruby>=0.23.0
|
|
30
|
+
Requires-Dist: tree-sitter-rust>=0.23.0
|
|
31
|
+
Requires-Dist: tree-sitter-typescript>=0.23.0
|
|
32
|
+
Requires-Dist: tree-sitter>=0.25.0
|
|
33
|
+
Provides-Extra: all
|
|
34
|
+
Requires-Dist: build; extra == 'all'
|
|
35
|
+
Requires-Dist: hatch; extra == 'all'
|
|
36
|
+
Requires-Dist: openai>=1.0.0; extra == 'all'
|
|
37
|
+
Requires-Dist: pytest>=7.0; extra == 'all'
|
|
38
|
+
Requires-Dist: twine; extra == 'all'
|
|
39
|
+
Provides-Extra: dev
|
|
40
|
+
Requires-Dist: build; extra == 'dev'
|
|
41
|
+
Requires-Dist: hatch; extra == 'dev'
|
|
42
|
+
Requires-Dist: pytest>=7.0; extra == 'dev'
|
|
43
|
+
Requires-Dist: twine; extra == 'dev'
|
|
44
|
+
Provides-Extra: llm
|
|
45
|
+
Requires-Dist: openai>=1.0.0; extra == 'llm'
|
|
46
|
+
Description-Content-Type: text/markdown
|
|
47
|
+
|
|
48
|
+
<div align="center">
|
|
49
|
+
|
|
50
|
+
<img src="docs/banner.svg" alt="okf-generator banner" width="100%"/>
|
|
51
|
+
|
|
52
|
+
<br/>
|
|
53
|
+
|
|
54
|
+
[](https://pypi.org/project/okf-generator/)
|
|
55
|
+
[](https://pypi.org/project/okf-generator/)
|
|
56
|
+
[](https://github.com/umairbaig/okf-generator/actions)
|
|
57
|
+
[](LICENSE)
|
|
58
|
+
[](https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing)
|
|
59
|
+
[](SKILL.md)
|
|
60
|
+
[](CONTRIBUTING.md)
|
|
61
|
+
|
|
62
|
+
**Index any codebase into a structured OKF v0.1 knowledge bundle — then look up exact concepts for AI agents like OpenCode.**
|
|
63
|
+
|
|
64
|
+
[Installation](#installation) · [Quick Start](#quick-start) · [CLI Reference](#cli-reference) · [OpenCode Integration](#opencode-integration) · [Contributing](#contributing)
|
|
65
|
+
|
|
66
|
+
</div>
|
|
67
|
+
|
|
68
|
+
---
|
|
69
|
+
|
|
70
|
+
## What is this?
|
|
71
|
+
|
|
72
|
+
`okf-generator` converts your source code into an [Open Knowledge Format](https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing) (OKF) v0.1 knowledge bundle — structured markdown files that AI agents can read, search, and reason over.
|
|
73
|
+
|
|
74
|
+
Instead of giving an AI your entire codebase, you give it exactly the concept it needs:
|
|
75
|
+
|
|
76
|
+
```bash
|
|
77
|
+
# Before touching WorldBankConnector, look it up
|
|
78
|
+
okf lookup WorldBankConnector
|
|
79
|
+
|
|
80
|
+
# CLASS: WorldBankConnector
|
|
81
|
+
# Source : StockAI/RnD/python/connectors/economic_data.py line 51
|
|
82
|
+
# Description : Fetches World Bank development indicators via wbdata API.
|
|
83
|
+
# Methods : get_indicator, search
|
|
84
|
+
# Signature : class WorldBankConnector
|
|
85
|
+
```
|
|
86
|
+
|
|
87
|
+
## Features
|
|
88
|
+
|
|
89
|
+
- **6 languages** — Python (stdlib AST), JS/TS/Go/Java/Rust/Ruby (tree-sitter)
|
|
90
|
+
- **Zero LLM required** for extraction — deterministic, fast, offline-capable
|
|
91
|
+
- **OKF v0.1 conformant** — type, description, resource, tags, timestamp
|
|
92
|
+
- **Domain/resource-path layout** — bundle mirrors your source tree exactly
|
|
93
|
+
- **Resumable LLM enrichment** — enrich descriptions with any OpenAI-compat endpoint; safe to interrupt and rerun
|
|
94
|
+
- **OpenCode integration** — `AGENTS.md` + custom commands for pinpoint context injection
|
|
95
|
+
- **Training data pipeline** — convert bundle to JSONL pairs (codegen, QA, doc, summarize, crosslink)
|
|
96
|
+
- **Claude Skill** — install `SKILL.md` to trigger the full pipeline from natural language
|
|
97
|
+
|
|
98
|
+
## Installation
|
|
99
|
+
|
|
100
|
+
```bash
|
|
101
|
+
# Core (extraction only — no LLM required)
|
|
102
|
+
pip install okf-generator
|
|
103
|
+
|
|
104
|
+
# With LLM enrichment + training pair generation
|
|
105
|
+
pip install okf-generator[llm]
|
|
106
|
+
```
|
|
107
|
+
|
|
108
|
+
**Requirements:** Python 3.11+
|
|
109
|
+
|
|
110
|
+
## Quick Start
|
|
111
|
+
|
|
112
|
+
```bash
|
|
113
|
+
# 1. Generate a knowledge bundle from your codebase
|
|
114
|
+
okf generate ./my_project ./okf_bundle
|
|
115
|
+
|
|
116
|
+
# 2. Look up a concept (works instantly, zero LLM)
|
|
117
|
+
okf lookup WorldBankConnector
|
|
118
|
+
|
|
119
|
+
# 3. Find all concepts from one file
|
|
120
|
+
okf lookup --file src/connectors/economic_data.py
|
|
121
|
+
|
|
122
|
+
# 4. Generate training pairs from the bundle
|
|
123
|
+
okf pairs ./okf_bundle ./train.jsonl
|
|
124
|
+
|
|
125
|
+
# 5. Regenerate SUMMARY.md after enrichment
|
|
126
|
+
okf summarize ./okf_bundle
|
|
127
|
+
```
|
|
128
|
+
|
|
129
|
+
## Bundle Layout
|
|
130
|
+
|
|
131
|
+
The output mirrors your source tree — not flat buckets:
|
|
132
|
+
|
|
133
|
+
```
|
|
134
|
+
okf_bundle/
|
|
135
|
+
├── SUMMARY.md ← bird's-eye view for AI agents
|
|
136
|
+
├── index.md ← root navigation
|
|
137
|
+
├── log.md ← generation history
|
|
138
|
+
└── StockAI/
|
|
139
|
+
└── RnD/
|
|
140
|
+
└── python/
|
|
141
|
+
└── connectors/
|
|
142
|
+
├── index.md ← lists all concepts in this folder
|
|
143
|
+
├── economic_data.md ← Module concept
|
|
144
|
+
└── economic_data/
|
|
145
|
+
├── WorldBankConnector.md ← Class
|
|
146
|
+
├── get_indicator.md ← Function
|
|
147
|
+
└── search.md ← Function
|
|
148
|
+
```
|
|
149
|
+
|
|
150
|
+
Each file is OKF v0.1 conformant:
|
|
151
|
+
|
|
152
|
+
```yaml
|
|
153
|
+
---
|
|
154
|
+
type: Class
|
|
155
|
+
title: WorldBankConnector
|
|
156
|
+
description: Fetches World Bank development indicators via wbdata API.
|
|
157
|
+
resource: StockAI/RnD/python/connectors/economic_data.py
|
|
158
|
+
tags:
|
|
159
|
+
- lang:python
|
|
160
|
+
- type:Class
|
|
161
|
+
- module:StockAI
|
|
162
|
+
- domain:RnD
|
|
163
|
+
- git:branch:main
|
|
164
|
+
- git:repo:TrainLLMs
|
|
165
|
+
timestamp: '2026-05-23T09:01:21Z'
|
|
166
|
+
---
|
|
167
|
+
|
|
168
|
+
# WorldBankConnector
|
|
169
|
+
|
|
170
|
+
...signature, docstring, params, returns, methods, related concepts...
|
|
171
|
+
```
|
|
172
|
+
|
|
173
|
+
## CLI Reference
|
|
174
|
+
|
|
175
|
+
### `okf generate`
|
|
176
|
+
|
|
177
|
+
```
|
|
178
|
+
okf generate <source_dir> [output_dir]
|
|
179
|
+
|
|
180
|
+
Options:
|
|
181
|
+
--summarize <bundle_dir> Regenerate SUMMARY.md only (no re-scan)
|
|
182
|
+
|
|
183
|
+
Environment variables (LLM enrichment):
|
|
184
|
+
OKF_ENRICH=1 Enable LLM enrichment
|
|
185
|
+
OKF_BASE_URL OpenAI-compat base URL (default: https://api.anthropic.com/v1)
|
|
186
|
+
OKF_API_KEY API key
|
|
187
|
+
OKF_MODEL Model name (default: claude-sonnet-4-6)
|
|
188
|
+
OKF_MAX_WORKERS Parallel workers (default: 2)
|
|
189
|
+
```
|
|
190
|
+
|
|
191
|
+
### `okf lookup`
|
|
192
|
+
|
|
193
|
+
```
|
|
194
|
+
okf lookup [query] [options]
|
|
195
|
+
|
|
196
|
+
Options:
|
|
197
|
+
--bundle PATH Bundle directory (default: ./okf_bundle)
|
|
198
|
+
--file PATH Filter by source file
|
|
199
|
+
--type TYPE Filter by concept type: Function | Class | Module
|
|
200
|
+
--tag TAG Filter by tag, repeatable: --tag lang:python
|
|
201
|
+
--limit N Max results (default: 10)
|
|
202
|
+
--compact One-line output per result
|
|
203
|
+
--json JSON output for programmatic use
|
|
204
|
+
--full Raw .md file content
|
|
205
|
+
--min-score N Minimum relevance score 0-1 (default: 0.1)
|
|
206
|
+
```
|
|
207
|
+
|
|
208
|
+
### `okf pairs`
|
|
209
|
+
|
|
210
|
+
```
|
|
211
|
+
okf pairs <bundle_dir> [output_file]
|
|
212
|
+
|
|
213
|
+
Environment variables:
|
|
214
|
+
SKIP_SYNTH=1 Static pairs only (no LLM)
|
|
215
|
+
SYNTH_BASE_URL API endpoint
|
|
216
|
+
SYNTH_API_KEY API key
|
|
217
|
+
SYNTH_MODEL Model name
|
|
218
|
+
MAX_WORKERS Parallel workers (default: 3)
|
|
219
|
+
QA_PER_CONCEPT Q&A pairs per concept (default: 3)
|
|
220
|
+
PAIR_TYPES Comma-separated: codegen,qa,doc,summarize,crosslink
|
|
221
|
+
```
|
|
222
|
+
|
|
223
|
+
## Supported Languages
|
|
224
|
+
|
|
225
|
+
| Language | Parser | Extracts |
|
|
226
|
+
|----------|--------|---------|
|
|
227
|
+
| Python | stdlib `ast` | Functions, classes, params, return types, docstrings |
|
|
228
|
+
| JavaScript / TypeScript | tree-sitter | Functions, arrow fns, classes, JSDoc |
|
|
229
|
+
| Go | tree-sitter | Funcs, methods, structs, interfaces, GoDoc |
|
|
230
|
+
| Java | tree-sitter | Classes, methods, constructors, Javadoc |
|
|
231
|
+
| Rust | tree-sitter | Fns, structs, enums, traits, impl blocks, `///` |
|
|
232
|
+
| Ruby | tree-sitter | Defs, classes, modules, `#` comments |
|
|
233
|
+
|
|
234
|
+
## LLM Enrichment
|
|
235
|
+
|
|
236
|
+
Works with any OpenAI-compatible endpoint — Claude, Ollama, llama.cpp, etc:
|
|
237
|
+
|
|
238
|
+
```bash
|
|
239
|
+
# Using a local llama.cpp server
|
|
240
|
+
OKF_ENRICH=1 \
|
|
241
|
+
OKF_BASE_URL="http://localhost:8080/v1" \
|
|
242
|
+
OKF_API_KEY="llamabarn" \
|
|
243
|
+
OKF_MODEL="ggml-org/gemma-3-4b-it-qat-GGUF:Q4_0" \
|
|
244
|
+
OKF_MAX_WORKERS=2 \
|
|
245
|
+
okf generate ./my_project ./okf_bundle
|
|
246
|
+
```
|
|
247
|
+
|
|
248
|
+
Enrichment is **resumable** — interrupt and rerun freely. Already-enriched concepts are skipped.
|
|
249
|
+
|
|
250
|
+
## OpenCode Integration
|
|
251
|
+
|
|
252
|
+
```bash
|
|
253
|
+
# 1. Tell OpenCode about the bundle (auto-loaded every session)
|
|
254
|
+
cat >> AGENTS.md << 'EOF'
|
|
255
|
+
## OKF Knowledge Bundle
|
|
256
|
+
Before working on any class or function, look it up:
|
|
257
|
+
okf lookup --bundle ./okf_bundle <ConceptName>
|
|
258
|
+
EOF
|
|
259
|
+
|
|
260
|
+
# 2. Add a custom command
|
|
261
|
+
mkdir -p .opencode/commands
|
|
262
|
+
echo "RUN okf lookup --bundle ./okf_bundle \$NAME" > .opencode/commands/lookup.md
|
|
263
|
+
```
|
|
264
|
+
|
|
265
|
+
Then in OpenCode: `/lookup NAME=WorldBankConnector`
|
|
266
|
+
|
|
267
|
+
See [docs/opencode-integration.md](references/opencode-integration.md) for full setup.
|
|
268
|
+
|
|
269
|
+
## Python API
|
|
270
|
+
|
|
271
|
+
```python
|
|
272
|
+
from okf.generator import scan_codebase, write_bundle, write_summary
|
|
273
|
+
from okf.lookup import load_bundle, search
|
|
274
|
+
|
|
275
|
+
# Generate bundle
|
|
276
|
+
concepts = scan_codebase("./my_project")
|
|
277
|
+
write_bundle(concepts, "./okf_bundle", "my_project", ["initial generation"])
|
|
278
|
+
write_summary("my_project", concepts, "./okf_bundle", {})
|
|
279
|
+
|
|
280
|
+
# Search concepts
|
|
281
|
+
bundle = load_bundle("./okf_bundle")
|
|
282
|
+
results = search(bundle, tokens=["WorldBankConnector"])
|
|
283
|
+
print(results[0]["description"])
|
|
284
|
+
```
|
|
285
|
+
|
|
286
|
+
## Training Data
|
|
287
|
+
|
|
288
|
+
Convert your OKF bundle into JSONL training pairs for fine-tuning:
|
|
289
|
+
|
|
290
|
+
```bash
|
|
291
|
+
# 5 pair types: codegen, qa, doc, summarize, crosslink
|
|
292
|
+
okf pairs ./okf_bundle ./train.jsonl
|
|
293
|
+
```
|
|
294
|
+
|
|
295
|
+
Each pair is in chat format compatible with most fine-tuning pipelines.
|
|
296
|
+
|
|
297
|
+
## Claude Skill
|
|
298
|
+
|
|
299
|
+
Install `SKILL.md` to trigger the full pipeline from natural language inside Claude:
|
|
300
|
+
|
|
301
|
+
> *"Index my codebase"* → generates OKF bundle
|
|
302
|
+
> *"Look up WorldBankConnector"* → returns exact concept
|
|
303
|
+
> *"Generate training pairs from my bundle"* → outputs JSONL
|
|
304
|
+
|
|
305
|
+
## Contributing
|
|
306
|
+
|
|
307
|
+
Contributions are welcome! See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
|
|
308
|
+
|
|
309
|
+
```bash
|
|
310
|
+
git clone https://github.com/umairbaig/okf-generator
|
|
311
|
+
cd okf-generator
|
|
312
|
+
pip install -e ".[dev]"
|
|
313
|
+
pytest tests/
|
|
314
|
+
```
|
|
315
|
+
|
|
316
|
+
**Good first issues:** adding a new language parser, improving fuzzy search scoring, adding a CHANGELOG.
|
|
317
|
+
|
|
318
|
+
## License
|
|
319
|
+
|
|
320
|
+
[MIT](LICENSE) — Copyright © 2026 Umair Baig
|
|
@@ -0,0 +1,273 @@
|
|
|
1
|
+
<div align="center">
|
|
2
|
+
|
|
3
|
+
<img src="docs/banner.svg" alt="okf-generator banner" width="100%"/>
|
|
4
|
+
|
|
5
|
+
<br/>
|
|
6
|
+
|
|
7
|
+
[](https://pypi.org/project/okf-generator/)
|
|
8
|
+
[](https://pypi.org/project/okf-generator/)
|
|
9
|
+
[](https://github.com/umairbaig/okf-generator/actions)
|
|
10
|
+
[](LICENSE)
|
|
11
|
+
[](https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing)
|
|
12
|
+
[](SKILL.md)
|
|
13
|
+
[](CONTRIBUTING.md)
|
|
14
|
+
|
|
15
|
+
**Index any codebase into a structured OKF v0.1 knowledge bundle — then look up exact concepts for AI agents like OpenCode.**
|
|
16
|
+
|
|
17
|
+
[Installation](#installation) · [Quick Start](#quick-start) · [CLI Reference](#cli-reference) · [OpenCode Integration](#opencode-integration) · [Contributing](#contributing)
|
|
18
|
+
|
|
19
|
+
</div>
|
|
20
|
+
|
|
21
|
+
---
|
|
22
|
+
|
|
23
|
+
## What is this?
|
|
24
|
+
|
|
25
|
+
`okf-generator` converts your source code into an [Open Knowledge Format](https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing) (OKF) v0.1 knowledge bundle — structured markdown files that AI agents can read, search, and reason over.
|
|
26
|
+
|
|
27
|
+
Instead of giving an AI your entire codebase, you give it exactly the concept it needs:
|
|
28
|
+
|
|
29
|
+
```bash
|
|
30
|
+
# Before touching WorldBankConnector, look it up
|
|
31
|
+
okf lookup WorldBankConnector
|
|
32
|
+
|
|
33
|
+
# CLASS: WorldBankConnector
|
|
34
|
+
# Source : StockAI/RnD/python/connectors/economic_data.py line 51
|
|
35
|
+
# Description : Fetches World Bank development indicators via wbdata API.
|
|
36
|
+
# Methods : get_indicator, search
|
|
37
|
+
# Signature : class WorldBankConnector
|
|
38
|
+
```
|
|
39
|
+
|
|
40
|
+
## Features
|
|
41
|
+
|
|
42
|
+
- **6 languages** — Python (stdlib AST), JS/TS/Go/Java/Rust/Ruby (tree-sitter)
|
|
43
|
+
- **Zero LLM required** for extraction — deterministic, fast, offline-capable
|
|
44
|
+
- **OKF v0.1 conformant** — type, description, resource, tags, timestamp
|
|
45
|
+
- **Domain/resource-path layout** — bundle mirrors your source tree exactly
|
|
46
|
+
- **Resumable LLM enrichment** — enrich descriptions with any OpenAI-compat endpoint; safe to interrupt and rerun
|
|
47
|
+
- **OpenCode integration** — `AGENTS.md` + custom commands for pinpoint context injection
|
|
48
|
+
- **Training data pipeline** — convert bundle to JSONL pairs (codegen, QA, doc, summarize, crosslink)
|
|
49
|
+
- **Claude Skill** — install `SKILL.md` to trigger the full pipeline from natural language
|
|
50
|
+
|
|
51
|
+
## Installation
|
|
52
|
+
|
|
53
|
+
```bash
|
|
54
|
+
# Core (extraction only — no LLM required)
|
|
55
|
+
pip install okf-generator
|
|
56
|
+
|
|
57
|
+
# With LLM enrichment + training pair generation
|
|
58
|
+
pip install okf-generator[llm]
|
|
59
|
+
```
|
|
60
|
+
|
|
61
|
+
**Requirements:** Python 3.11+
|
|
62
|
+
|
|
63
|
+
## Quick Start
|
|
64
|
+
|
|
65
|
+
```bash
|
|
66
|
+
# 1. Generate a knowledge bundle from your codebase
|
|
67
|
+
okf generate ./my_project ./okf_bundle
|
|
68
|
+
|
|
69
|
+
# 2. Look up a concept (works instantly, zero LLM)
|
|
70
|
+
okf lookup WorldBankConnector
|
|
71
|
+
|
|
72
|
+
# 3. Find all concepts from one file
|
|
73
|
+
okf lookup --file src/connectors/economic_data.py
|
|
74
|
+
|
|
75
|
+
# 4. Generate training pairs from the bundle
|
|
76
|
+
okf pairs ./okf_bundle ./train.jsonl
|
|
77
|
+
|
|
78
|
+
# 5. Regenerate SUMMARY.md after enrichment
|
|
79
|
+
okf summarize ./okf_bundle
|
|
80
|
+
```
|
|
81
|
+
|
|
82
|
+
## Bundle Layout
|
|
83
|
+
|
|
84
|
+
The output mirrors your source tree — not flat buckets:
|
|
85
|
+
|
|
86
|
+
```
|
|
87
|
+
okf_bundle/
|
|
88
|
+
├── SUMMARY.md ← bird's-eye view for AI agents
|
|
89
|
+
├── index.md ← root navigation
|
|
90
|
+
├── log.md ← generation history
|
|
91
|
+
└── StockAI/
|
|
92
|
+
└── RnD/
|
|
93
|
+
└── python/
|
|
94
|
+
└── connectors/
|
|
95
|
+
├── index.md ← lists all concepts in this folder
|
|
96
|
+
├── economic_data.md ← Module concept
|
|
97
|
+
└── economic_data/
|
|
98
|
+
├── WorldBankConnector.md ← Class
|
|
99
|
+
├── get_indicator.md ← Function
|
|
100
|
+
└── search.md ← Function
|
|
101
|
+
```
|
|
102
|
+
|
|
103
|
+
Each file is OKF v0.1 conformant:
|
|
104
|
+
|
|
105
|
+
```yaml
|
|
106
|
+
---
|
|
107
|
+
type: Class
|
|
108
|
+
title: WorldBankConnector
|
|
109
|
+
description: Fetches World Bank development indicators via wbdata API.
|
|
110
|
+
resource: StockAI/RnD/python/connectors/economic_data.py
|
|
111
|
+
tags:
|
|
112
|
+
- lang:python
|
|
113
|
+
- type:Class
|
|
114
|
+
- module:StockAI
|
|
115
|
+
- domain:RnD
|
|
116
|
+
- git:branch:main
|
|
117
|
+
- git:repo:TrainLLMs
|
|
118
|
+
timestamp: '2026-05-23T09:01:21Z'
|
|
119
|
+
---
|
|
120
|
+
|
|
121
|
+
# WorldBankConnector
|
|
122
|
+
|
|
123
|
+
...signature, docstring, params, returns, methods, related concepts...
|
|
124
|
+
```
|
|
125
|
+
|
|
126
|
+
## CLI Reference
|
|
127
|
+
|
|
128
|
+
### `okf generate`
|
|
129
|
+
|
|
130
|
+
```
|
|
131
|
+
okf generate <source_dir> [output_dir]
|
|
132
|
+
|
|
133
|
+
Options:
|
|
134
|
+
--summarize <bundle_dir> Regenerate SUMMARY.md only (no re-scan)
|
|
135
|
+
|
|
136
|
+
Environment variables (LLM enrichment):
|
|
137
|
+
OKF_ENRICH=1 Enable LLM enrichment
|
|
138
|
+
OKF_BASE_URL OpenAI-compat base URL (default: https://api.anthropic.com/v1)
|
|
139
|
+
OKF_API_KEY API key
|
|
140
|
+
OKF_MODEL Model name (default: claude-sonnet-4-6)
|
|
141
|
+
OKF_MAX_WORKERS Parallel workers (default: 2)
|
|
142
|
+
```
|
|
143
|
+
|
|
144
|
+
### `okf lookup`
|
|
145
|
+
|
|
146
|
+
```
|
|
147
|
+
okf lookup [query] [options]
|
|
148
|
+
|
|
149
|
+
Options:
|
|
150
|
+
--bundle PATH Bundle directory (default: ./okf_bundle)
|
|
151
|
+
--file PATH Filter by source file
|
|
152
|
+
--type TYPE Filter by concept type: Function | Class | Module
|
|
153
|
+
--tag TAG Filter by tag, repeatable: --tag lang:python
|
|
154
|
+
--limit N Max results (default: 10)
|
|
155
|
+
--compact One-line output per result
|
|
156
|
+
--json JSON output for programmatic use
|
|
157
|
+
--full Raw .md file content
|
|
158
|
+
--min-score N Minimum relevance score 0-1 (default: 0.1)
|
|
159
|
+
```
|
|
160
|
+
|
|
161
|
+
### `okf pairs`
|
|
162
|
+
|
|
163
|
+
```
|
|
164
|
+
okf pairs <bundle_dir> [output_file]
|
|
165
|
+
|
|
166
|
+
Environment variables:
|
|
167
|
+
SKIP_SYNTH=1 Static pairs only (no LLM)
|
|
168
|
+
SYNTH_BASE_URL API endpoint
|
|
169
|
+
SYNTH_API_KEY API key
|
|
170
|
+
SYNTH_MODEL Model name
|
|
171
|
+
MAX_WORKERS Parallel workers (default: 3)
|
|
172
|
+
QA_PER_CONCEPT Q&A pairs per concept (default: 3)
|
|
173
|
+
PAIR_TYPES Comma-separated: codegen,qa,doc,summarize,crosslink
|
|
174
|
+
```
|
|
175
|
+
|
|
176
|
+
## Supported Languages
|
|
177
|
+
|
|
178
|
+
| Language | Parser | Extracts |
|
|
179
|
+
|----------|--------|---------|
|
|
180
|
+
| Python | stdlib `ast` | Functions, classes, params, return types, docstrings |
|
|
181
|
+
| JavaScript / TypeScript | tree-sitter | Functions, arrow fns, classes, JSDoc |
|
|
182
|
+
| Go | tree-sitter | Funcs, methods, structs, interfaces, GoDoc |
|
|
183
|
+
| Java | tree-sitter | Classes, methods, constructors, Javadoc |
|
|
184
|
+
| Rust | tree-sitter | Fns, structs, enums, traits, impl blocks, `///` |
|
|
185
|
+
| Ruby | tree-sitter | Defs, classes, modules, `#` comments |
|
|
186
|
+
|
|
187
|
+
## LLM Enrichment
|
|
188
|
+
|
|
189
|
+
Works with any OpenAI-compatible endpoint — Claude, Ollama, llama.cpp, etc:
|
|
190
|
+
|
|
191
|
+
```bash
|
|
192
|
+
# Using a local llama.cpp server
|
|
193
|
+
OKF_ENRICH=1 \
|
|
194
|
+
OKF_BASE_URL="http://localhost:8080/v1" \
|
|
195
|
+
OKF_API_KEY="llamabarn" \
|
|
196
|
+
OKF_MODEL="ggml-org/gemma-3-4b-it-qat-GGUF:Q4_0" \
|
|
197
|
+
OKF_MAX_WORKERS=2 \
|
|
198
|
+
okf generate ./my_project ./okf_bundle
|
|
199
|
+
```
|
|
200
|
+
|
|
201
|
+
Enrichment is **resumable** — interrupt and rerun freely. Already-enriched concepts are skipped.
|
|
202
|
+
|
|
203
|
+
## OpenCode Integration
|
|
204
|
+
|
|
205
|
+
```bash
|
|
206
|
+
# 1. Tell OpenCode about the bundle (auto-loaded every session)
|
|
207
|
+
cat >> AGENTS.md << 'EOF'
|
|
208
|
+
## OKF Knowledge Bundle
|
|
209
|
+
Before working on any class or function, look it up:
|
|
210
|
+
okf lookup --bundle ./okf_bundle <ConceptName>
|
|
211
|
+
EOF
|
|
212
|
+
|
|
213
|
+
# 2. Add a custom command
|
|
214
|
+
mkdir -p .opencode/commands
|
|
215
|
+
echo "RUN okf lookup --bundle ./okf_bundle \$NAME" > .opencode/commands/lookup.md
|
|
216
|
+
```
|
|
217
|
+
|
|
218
|
+
Then in OpenCode: `/lookup NAME=WorldBankConnector`
|
|
219
|
+
|
|
220
|
+
See [docs/opencode-integration.md](references/opencode-integration.md) for full setup.
|
|
221
|
+
|
|
222
|
+
## Python API
|
|
223
|
+
|
|
224
|
+
```python
|
|
225
|
+
from okf.generator import scan_codebase, write_bundle, write_summary
|
|
226
|
+
from okf.lookup import load_bundle, search
|
|
227
|
+
|
|
228
|
+
# Generate bundle
|
|
229
|
+
concepts = scan_codebase("./my_project")
|
|
230
|
+
write_bundle(concepts, "./okf_bundle", "my_project", ["initial generation"])
|
|
231
|
+
write_summary("my_project", concepts, "./okf_bundle", {})
|
|
232
|
+
|
|
233
|
+
# Search concepts
|
|
234
|
+
bundle = load_bundle("./okf_bundle")
|
|
235
|
+
results = search(bundle, tokens=["WorldBankConnector"])
|
|
236
|
+
print(results[0]["description"])
|
|
237
|
+
```
|
|
238
|
+
|
|
239
|
+
## Training Data
|
|
240
|
+
|
|
241
|
+
Convert your OKF bundle into JSONL training pairs for fine-tuning:
|
|
242
|
+
|
|
243
|
+
```bash
|
|
244
|
+
# 5 pair types: codegen, qa, doc, summarize, crosslink
|
|
245
|
+
okf pairs ./okf_bundle ./train.jsonl
|
|
246
|
+
```
|
|
247
|
+
|
|
248
|
+
Each pair is in chat format compatible with most fine-tuning pipelines.
|
|
249
|
+
|
|
250
|
+
## Claude Skill
|
|
251
|
+
|
|
252
|
+
Install `SKILL.md` to trigger the full pipeline from natural language inside Claude:
|
|
253
|
+
|
|
254
|
+
> *"Index my codebase"* → generates OKF bundle
|
|
255
|
+
> *"Look up WorldBankConnector"* → returns exact concept
|
|
256
|
+
> *"Generate training pairs from my bundle"* → outputs JSONL
|
|
257
|
+
|
|
258
|
+
## Contributing
|
|
259
|
+
|
|
260
|
+
Contributions are welcome! See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
|
|
261
|
+
|
|
262
|
+
```bash
|
|
263
|
+
git clone https://github.com/umairbaig/okf-generator
|
|
264
|
+
cd okf-generator
|
|
265
|
+
pip install -e ".[dev]"
|
|
266
|
+
pytest tests/
|
|
267
|
+
```
|
|
268
|
+
|
|
269
|
+
**Good first issues:** adding a new language parser, improving fuzzy search scoring, adding a CHANGELOG.
|
|
270
|
+
|
|
271
|
+
## License
|
|
272
|
+
|
|
273
|
+
[MIT](LICENSE) — Copyright © 2026 Umair Baig
|