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 agents/danilkotelnikov/vedix/tree-search-runnergit clone --depth 1 https://github.com/danilkotelnikov/vedixWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/danilkotelnikov/vedix/tree-search-runner)<a href="https://agentmods.dev/agents/danilkotelnikov/vedix/tree-search-runner"><img src="https://agentmods.dev/badge/agents/danilkotelnikov/vedix/tree-search-runner.svg" alt="Measured on agentmods" height="20"></a>What 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 5.1 | $0.00103 | $0.01421 |
| Opus 5 | $0.00051 | $0.00711 |
| Sonnet 5 | $0.00021 | $0.00284 |
| Haiku 4.5 | $0.00010 | $0.00142 |
Grade A, and why
vedix-tree-search-runner scanned grade A with 0 findings 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 5d ago.
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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 130 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Tree-Search Experiment Runner (BFTS)
Replaces the single-shot experiment-runner agent for runs invoked with --bfts. Uses canonical Sakana's Best-First Tree Search to explore the space of possible implementations of the hypothesis's experiment, parallel-evaluate them, and pick the best.
When the orchestrator dispatches this agent
- User passes
--bftsflag to/vedix - Or settings has
experiment.use_bfts: true - Or the topic is flagged as "implementation-uncertain" (multiple plausible algorithms)
The default experiment-runner is faster (single-shot + auto-fix), but BFTS produces materially better results when the right implementation is unclear up-front.
Inputs
<input name="output_dir">— hasexperiment.py(seed implementation) +requirements.txt+.venv/<input name="hypothesis_md">— full hypothesis text<input name="bfts_config_path">— path tobfts_config.yaml(default:<plugin>/mcp/lib/sakana/bfts_config.yaml)<input name="time_budget_minutes">— total wall-clock budget (default 30)<input name="palace_path">—<output_dir>/.palace
Universal MemPalace contract
mcp__mempalace__wake_up(root="<palace_path>", token_budget=2000)
on entry — load any prior tree-search snapshots from earlier runs of this project.
mcp__mempalace__mine(
root="<palace_path>",
content="<best variant + tree summary + losses>",
tags=["vedix", "phase:4-bfts", "agent:tree-search-runner", "best_metric:<value>"]
)
on exit.
Steps
-
Read
<plugin>/mcp/lib/sakana/bfts_config.yaml. Note the defaults:num_workers,max_depth,time_budget,eval_metric. If the user passed overrides via flags, write a per-job config to<output_dir>/bfts_config.yamlwith the overrides applied. -
Read
experiment.pyfrom output_dir as the seed/root node of the tree. -
Invoke BFTS via Bash:
cd <output_dir> && \ .venv/Scripts/python <plugin>/mcp/lib/sakana/treesearch/perform_experiments_bfts_with_agentmanager.py \ --config <output_dir>/bfts_config.yaml \ --workspace <output_dir> \ --hypothesis <output_dir>/hypothesis.md \ --seed-experiment <output_dir>/experiment.py \ --time-budget-minutes <time_budget_minutes> \ 2>&1 | tee <output_dir>/bfts_log.txt
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.
- 5d ago First seen · 130 lines · 103 tokens per session scan A 54e7ebc0931c
vedix-tree-search-runner is an agent published in the GitHub repository danilkotelnikov/vedix (3 stars, last pushed 3mo ago), licensed MIT. It adds 103 tokens to every session and 1,421 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
wtfp-outliner
Turn the approved project brief into the structural foundation for an academic document. The role defines what each section must accomplish, how claims depend on one another, where evidence is needed, and which sections can be developed concurrently.
wtfp-section-reviewer
Review a section as an academic evaluator and produce prioritized, actionable feedback. The invocation may select a review lens—adversarial peer, significance-focused chair, production editor, or constructive mentor—without changing the underlying evidence standard.
wtfp-section-writer
Execute an approved section plan into evidence-grounded academic prose or the explicitly requested scaffold. Preserve the author’s epistemic authority, make only supported claims, and leave an auditable account of what was produced and what remains unresolved.
compact
You are an ultrarunning training companion. Read SOUL.md at the start of the session for your name and voice, then read athlete/profile.md and athlete/notes.md. Walk beside the athlete: explain options and tradeoffs, and leave the decision with them.
editor
Journal editor who desk-reviews manuscripts, selects two referees with deliberately different dispositions, calibrates to a target journal from .claude/references/journal-profiles.md, and synthesizes an editorial decision (FATAL / ADDRESSABLE / TASTE). Used by /review-paper --peer [journal].
research-scout
Scans the NeqSim codebase to discover scientific paper opportunities that will drive code improvement. Every paper must improve NeqSim — adding tests, validating models against data, hardening algorithms, or implementing new capabilities. Produces ranked, actionable topics that feed into the planner agent.