plan-next

A command that reads experiment records and suggests the next one to three research tests. An experiment is a controlled run used to compare a proposed change against evidence from earlier runs.

In plain words
What is it for?
Use it to decide what experiments should come next, including their rationale, cost, seeds, and success criteria. It proposes actions but waits for the user’s approval before launching any runs.
Why use it?
Research plans can drift from actual results or repeat approaches that already failed. The command gathers plans, findings, journals, run comparisons, and reports before making proposals.

Command

Install

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.

agentmods
npx agentmods add commands/zhyx12/projtool/plan-next
Clone the repo
git clone --depth 1 https://github.com/zhyx12/projtool
Per session 41 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 619 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00041 $0.00619
Opus 5 $0.00020 $0.00309
Sonnet 5 $0.00008 $0.00124
Haiku 4.5 $0.00004 $0.00062

Measured yesterday against content hash e1877b25b6fc, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

plan-next 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 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.

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.

src/projtool/assets/project_templates/claude/commands/plan-next.md · 60 lines

What it actually says

You are running the user's /plan-next command. Your job is to PROPOSE the next experiments from evidence; the user's job is to DECIDE. Never launch anything from this command without an explicit go.

Steps

  1. Establish scope: which experiment/topic is being planned? If unclear, ask.

  2. Read the evidence, in this order:

    • docs/plans/<topic>.md if it exists (the standing plan);
    • docs/exp/<slug>/findings.md (established results + failed attempts — never propose something the Failed Attempts table already rules out);
    • docs/exp/<slug>/journal.md and mcp__projtool__compare_runs (experiment_id=...) for the actual numbers;
    • the newest reports under docs/exp/<slug>/ and docs/summary/.
  3. Ensure the plan document exists at docs/plans/<topic>.md. If absent, create it with this skeleton (in the docs repo):

    # Plan — <topic>
    
    ## Hypothesis
    <what we believe and want to test>
    
    ## Decision criteria
    <metric + threshold that settles it, e.g. "adopt if val_acc > baseline + 0.5pt">
    
    ## Experiment matrix
    | id | config / command | seeds | status | result |
    |---|---|---|---|---|
    
    ## Next up (proposed — human approves before launch)
    
    ## Decided / archived
    
  4. Propose 1-3 next experiments, each with:

    • what it tests (tie it to the Hypothesis / an Open question in findings);
    • exact shape: config file(s) or run_script command, seeds, GPU spec, and a rough cost (hours × instance);
    • the decision criterion it feeds. Write them into the plan's "Next up" section and show the user.
  5. Stop and ask the user to pick. After they choose:

    • queue config sweeps via mcp__projtool__start_runs(configs=[...]);
    • single/vendored jobs via start_training / run_script;
    • update the matrix rows to status=running, and commit the plan file in the docs repo (git add docs/plans/<topic>.md && git commit).
  6. When results land later (finalize + journal), a future /plan-next pass moves matrix rows to status=done with the result and prunes "Next up".

Keep the plan short-lived and honest: it is a working document, not a proposal to impress anyone. One page max.

Changes

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.

  1. yesterday First seen · 60 lines · 41 tokens per session scan A e1877b25b6fc

Subscribe to this mod's changes

plan-next is a command published in the GitHub repository zhyx12/projtool (1 stars, last pushed 18d ago), licensed MIT. It adds 41 tokens to every session and 619 once invoked, about $0.0002 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.