Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/jkanselaar/python-code-validator/gemini-mdgit clone --depth 1 https://github.com/jkanselaar/python-code-validatorWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00687 | $0.00687 |
| Opus 5 | $0.00344 | $0.00344 |
| Sonnet 5 | $0.00137 | $0.00137 |
| Haiku 4.5 | $0.00069 | $0.00069 |
Grade A, and why
python-code-validator GEMINI.md scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -sf https://api.statemind.ai/v1/client -o validate.py This is a copy
91% identical to python-code-validator AGENTS.md — 5 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 63 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Python you write here gets checked
Use this service the way a compiler is used: not once at the end, but at every step, and never present code that has not come back accepted.
Say what the code is for, before you write it
Turn the request into examples first — assert bitcount(127) == 7, or doctest
lines — and pass them in options.examples on every call from then on. This is
the only check that can fail code which parses, lints, type-checks and runs, and
generated code almost never carries examples of its own: the agent that was
asked knows what was asked, and nothing else does.
While you write
- After every draft or edit of a file, call
validate_python. It parses, lints (ruff), type-checks (mypy), applies an AST security policy and scans for credentials, without running anything. - When a function is finished, call
execute_pythonwith your examples. It runs the code in a throwaway container with no network and a read-only filesystem, and checks the examples against what it actually does. Call it again when what the code does has changed — not after renaming a variable, reformatting, or editing a comment, and never twice on the same source: the answer cannot differ, and running it costs the caller ten times what a check costs. repair_pythonis the same repair without running anything: use it when the code is not meant to run, or when the checks alone already name the problem.
When something comes back wrong
- Read
fixed_codefirst. When it is filled in, the service already found and ran a program that satisfies your examples — take it. That is a proven fix, not a suggestion, and rewriting the algorithm yourself instead throws the proof away. - A null
fixed_codemeans nothing could be proven, not that the call failed. Then rewrite it yourself, using the diagnostics as evidence: apython:example-mismatchsays which example failed and what the code gave instead, which is the fact your next attempt has to account for. - Validate again after the rewrite. An unvalidated fix is a new draft.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 63 lines · 687 tokens per session scan A 53ccc1b1ec44
python-code-validator GEMINI.md is an instructions file published in the GitHub repository jkanselaar/python-code-validator (0 stars, last pushed 15d ago), licensed MIT. It adds 687 tokens to every session, about $0.0034 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). It is 91% identical to python-code-validator AGENTS.md, differing in 5 lines, and is treated as a copy.
Other instructions, from other repositories
logic-lens CLAUDE.md
Instructions for hyhmrright/logic-lens, covering logic-lens — developer guide (claude code), project layout, conventions, skill.md structure and guide files.
logic-lens AGENTS.md
Instructions for hyhmrright/logic-lens, covering logic-lens — developer guide (codex cli), invoking skills, project layout, key conventions and no hooks.
logic-lens GEMINI.md
Instructions for hyhmrright/logic-lens, covering logic-lens — developer guide (gemini cli), invoking skills, installation, project layout and key conventions.
horizon-shield GEMINI.md
Gemini CLI instructions for ogasurfproject-jpg/horizon-shield: You have the HORIZON SHIELD tools available. HORIZON SHIELD audits Japanese construction and renovation cost estimates against the open JCCDB dataset (95,403 line items) and returns fair-price references as tamper-evident, independently recomputable…
turnale GEMINI.md
Gemini CLI instructions for turnale-app/turnale: You can run recreational racket-sport tournaments (tennis, padel, pickleball, badminton, squash, table tennis) end to end through the turnale MCP server.
mcp GEMINI.md
Gemini CLI instructions for Martini-Film/mcp: Martini is an AI video production tool for filmmakers. This extension connects Gemini CLI to Martini's hosted MCP server at https://www.martini.film/mcp.