monty-codeact

Name: monty-codeact
Author: jsturtevant

34stars9forks

How to install monty-codeact?

jsturtevant/copilot-codeact-plugin/monty-codeact

$npx -y skills add jsturtevant/copilot-codeact-plugin --skill monty-codeact

Installs into the current project.

Summary

CodeAct via Monty. Use when looping over many files (8+), cross-referencing results from multiple sources, aggregating data across directories, or chaining 5+ dependent tool calls. Collapses N round-trips into one sandboxed Python run via scripts/codeact.py. NOT beneficial for <5

Files · 1

View on GitHub

SKILL.md

1	# Monty CodeAct
2
3	Collapse multi-step tool chains into a single sandboxed Python execution using
4	[Pydantic Monty](https://github.com/pydantic/monty) — a minimal, secure Python
5	interpreter written in Rust with sub-microsecond startup.
6
7	## Syntax
8
9	Tools are called as regular Python functions with keyword arguments:
10
11	```python
12	content = view(path="src/main.py")
13	files = glob(pattern="*/.py")
14	hits = grep(pattern="TODO", paths="src")
15	result = bash(command="git log --oneline -5")
16	edit(path="config.json", old_str='"debug": false', new_str='"debug": true')
17	```
18
19	Chain by sequencing:
20
21	```python
22	for f in glob(pattern="*/.py", paths="src"):
23	content = view(path=f)
24	if "TODO" in content:
25	print(f + ": " + str(content.count("TODO")) + " TODOs")
26	```
27
28	## Discover tools
29
30	```bash
31	python3 scripts/codeact.py --discover # JSON manifest
32	python3 scripts/codeact.py --instructions # LLM-ready reference
33	```
34
35	Tools are auto-detected based on what's installed on the host.
36
37	## Execute
38
39	```bash
40	uv run --with pydantic-monty python3 scripts/codeact.py --auto --workspace . --code '...'
41	```
42
43	Output: `{"stdout": "...", "stderr": "...", "return_value": null, "success": true}`
44
45	## Monty limitations
46
47	Monty runs a subset of Python. Will error on:
48	- Classes — no `class` keyword at all
49	- Match statements — no `match`/`case`
50	- f-string format specs — `f"{x:<10}"`, `f"{x:>5}"`, `f"{x:.2f}"` all fail
51	- `str.format()` — `"{:<10}".format(x)` fails
52	- `str.startswith()` with tuple — use `or` instead
53	- Set comprehensions — build with list + `in` checks
54	- Third-party imports — only stdlib subset
55	- Most stdlib — only: json, re, datetime, sys, os.environ (no os.path, no os.walk)
56	- Brace expansion in glob — `glob(pattern="src/{db,services}/*/.py")` is supported.
57
58	Sandbox tool return types (getting these wrong causes retries):
59	- `glob(pattern=...)` → list of strings like `["src/app.py", "src/utils.py"]`
60	- `view(path=...)` → string (file content)
61	- `mcp_call(server=..., tool=..., ...)` → string
62	- `bash(command=...)` → dict with keys `stdout`, `stderr`, `returncode`
63
64	Key usage patterns:
65	- Do NOT scout first. `glob()` and `view()` are in the sandbox — discover
66	files inside your codeact program, not with separate tool calls before it.
67	- One program, one bash call. Do not run multiple codeact invocations.
68	If the first one fails, fix the bug in the program, don't add a scouting step.
69	- Wrap file reads in try/except so one bad file doesn't abort the run.
70	- Use `for f in glob(pattern="*/.py"):` to iterate files. No os.walk or os.path.
71
72	Output formatting workaround (use instead of format specs):
73	```python
74	def pad(s, w):
75	s = str(s)
76	return s + " " * max(0, w - len(s))
77	```
78
79	Tips: Use `chr(10)` for newlines. Use `import json` explicitly.
80	Use string concatenation (`+`) or simple f-strings (`f"count: {n}"`).
81
82	## Trust model
83
84	Sandboxed code can only reach the outside world through registered tool
85	functions. Use `--workspace` to restrict file tools to a directory tree.
86
87	## Prerequisites
88
89	- Python 3.10+
90	- `uv` (recommended) or `pip install pydantic-monty`
91
92	## References
93
94	- [references/tool-patterns.md](references/tool-patterns.md)

People who liked this also viewed

6 picks

Type

monty-codeact

34stars9forks

View on GitHub

How to install monty-codeact?

jsturtevant/copilot-codeact-plugin/monty-codeact

$npx -y skills add jsturtevant/copilot-codeact-plugin --skill monty-codeact

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Monty CodeAct
2
3	Collapse multi-step tool chains into a single sandboxed Python execution using
4	[Pydantic Monty](https://github.com/pydantic/monty) — a minimal, secure Python
5	interpreter written in Rust with sub-microsecond startup.
6
7	## Syntax
8
9	Tools are called as regular Python functions with keyword arguments:
10
11	```python
12	content = view(path="src/main.py")
13	files = glob(pattern="*/.py")
14	hits = grep(pattern="TODO", paths="src")
15	result = bash(command="git log --oneline -5")
16	edit(path="config.json", old_str='"debug": false', new_str='"debug": true')
17	```
18
19	Chain by sequencing:
20
21	```python
22	for f in glob(pattern="*/.py", paths="src"):
23	content = view(path=f)
24	if "TODO" in content:
25	print(f + ": " + str(content.count("TODO")) + " TODOs")
26	```
27
28	## Discover tools
29
30	```bash
31	python3 scripts/codeact.py --discover # JSON manifest
32	python3 scripts/codeact.py --instructions # LLM-ready reference
33	```
34
35	Tools are auto-detected based on what's installed on the host.
36
37	## Execute
38
39	```bash
40	uv run --with pydantic-monty python3 scripts/codeact.py --auto --workspace . --code '...'
41	```
42
43	Output: `{"stdout": "...", "stderr": "...", "return_value": null, "success": true}`
44
45	## Monty limitations
46
47	Monty runs a subset of Python. Will error on:
48	- Classes — no `class` keyword at all
49	- Match statements — no `match`/`case`
50	- f-string format specs — `f"{x:<10}"`, `f"{x:>5}"`, `f"{x:.2f}"` all fail
51	- `str.format()` — `"{:<10}".format(x)` fails
52	- `str.startswith()` with tuple — use `or` instead
53	- Set comprehensions — build with list + `in` checks
54	- Third-party imports — only stdlib subset
55	- Most stdlib — only: json, re, datetime, sys, os.environ (no os.path, no os.walk)
56	- Brace expansion in glob — `glob(pattern="src/{db,services}/*/.py")` is supported.
57
58	Sandbox tool return types (getting these wrong causes retries):
59	- `glob(pattern=...)` → list of strings like `["src/app.py", "src/utils.py"]`
60	- `view(path=...)` → string (file content)
61	- `mcp_call(server=..., tool=..., ...)` → string
62	- `bash(command=...)` → dict with keys `stdout`, `stderr`, `returncode`
63
64	Key usage patterns:
65	- Do NOT scout first. `glob()` and `view()` are in the sandbox — discover
66	files inside your codeact program, not with separate tool calls before it.
67	- One program, one bash call. Do not run multiple codeact invocations.
68	If the first one fails, fix the bug in the program, don't add a scouting step.
69	- Wrap file reads in try/except so one bad file doesn't abort the run.
70	- Use `for f in glob(pattern="*/.py"):` to iterate files. No os.walk or os.path.
71
72	Output formatting workaround (use instead of format specs):
73	```python
74	def pad(s, w):
75	s = str(s)
76	return s + " " * max(0, w - len(s))
77	```
78
79	Tips: Use `chr(10)` for newlines. Use `import json` explicitly.
80	Use string concatenation (`+`) or simple f-strings (`f"count: {n}"`).
81
82	## Trust model
83
84	Sandboxed code can only reach the outside world through registered tool
85	functions. Use `--workspace` to restrict file tools to a directory tree.
86
87	## Prerequisites
88
89	- Python 3.10+
90	- `uv` (recommended) or `pip install pydantic-monty`
91
92	## References
93
94	- [references/tool-patterns.md](references/tool-patterns.md)

People who liked this also viewed

6 picks

Type