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/kastalien-research/thoughtbox/agents-mdgit clone --depth 1 https://github.com/Kastalien-Research/thoughtboxWrote 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/instructions/kastalien-research/thoughtbox/agents-md)<a href="https://agentmods.dev/instructions/kastalien-research/thoughtbox/agents-md"><img src="https://agentmods.dev/badge/instructions/kastalien-research/thoughtbox/agents-md.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 | $0.03222 | $0.03222 |
| Opus 5 | $0.01611 | $0.01611 |
| Sonnet 5 | $0.00644 | $0.00644 |
| Haiku 4.5 | $0.00322 | $0.00322 |
Grade C, and why
thoughtbox AGENTS.md scanned grade C 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 today.
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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- `PreToolUse` / `BeforeTool`: apply command safety checks before running risky shell commands. Block direct pushes to protected branches, force pushes, branch deletion, dangerous `rm -rf`, and unrequested writes to `.en How it starts
The opening of the file, as written. The whole thing — 226 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Scope
This file is authoritative for the MCP server, plugin, hub, and repo-level tooling. For web app conventions (Next.js 15 app under apps/web/), see apps/web/AGENTS.md.
Development Workflow (Source of Truth)
Every unit of work runs through a mandatory skill. Two additional protocols activate conditionally.
Mandatory Skills
workflow — Full development lifecycle from ideation through merge. Stage 2 produces or updates .specs/ markdown with YAML frontmatter claims (not ADRs). PR claims reference specs as spec_id:claim_id in prs/<branch>.json.
Read .claude/skills/workflow/SKILL.md for multi-stage feature work. For architectural decisions, update the relevant spec with frontmatter claims and an acceptance/evidence plan before implementation.
Conditional Protocols
ulysses-protocol — Activates when 2 consecutive surprises occur during a task. A surprise-gated debugging framework that prevents hallucinated progress through pre-committed recovery actions and falsifiable hypotheses.
Invoke only when stuck: .claude/skills/ulysses-protocol/SKILL.md
theseus-protocol — Use when the task is a refactor (structure changes, behavior preserved). Prevents Refactoring Fugue State via scope locking, adversarial Cassandra audits, and hard reversibility. Do not use for feature work or bug fixes.
Invoke only for refactoring tasks: .claude/skills/theseus-protocol/SKILL.md
Key Rules (always apply)
- Specs go in
.specs/(notspecs/). Active authority is spec markdown with YAML frontmatter claims (see.schemas/spec-v1.json). Archived ADRs live underdocs/decisions/archive/for historical context only. - Code and spec updates in the same commit. If you change code that a spec describes, update the spec in the same commit.
- Atomic commits. One sub-agent = one unit of work = one commit, made after review validates the work.
- Sub-agent summaries state: Claims, Spec/Evidence Alignment, Tests run, Known Gaps, Risks.
- Default: human is NOT in the loop. Operate autonomously. Escalate only when genuinely stuck after investigation.
- Orchestrators don't do manual work. Deploy sub-agents or agent teams. Protect your context window.
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.
- today First seen · 226 lines · 3,222 tokens per session scan C 928355fd20d5
thoughtbox AGENTS.md is an instructions file published in the GitHub repository Kastalien-Research/thoughtbox (64 stars, last pushed 1mo ago), licensed MIT. It adds 3,222 tokens to every session, about $0.0161 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other instructions, from other repositories
llm-context.py CLAUDE.md
Instructions for cyberchitta/llm-context.py, covering claude.md, working notes (gitignored) and draining the field notes.
seekstone CLAUDE.md
Instructions for shaqmughal/seekstone, covering claude.md, what this repo is, commands, the harness itself (run after npm install) and architecture.
cyxj-groksearch CLAUDE.md
Claude Code instructions for chenyuxiaojin/cyxj-groksearch, covering claude.md, 常用命令 and 结构与坑.
mcpelevator CLAUDE.md
Instructions for pacnpal/mcpelevator, covering commands, architecture, conventions, agent skills and issue tracker.
inkwell-memory CLAUDE.md
Instructions for veronchenko/inkwell-memory, covering claude.md — inkwellmemory, layout, multi-tenant mode (inkwellmultitenant=1), conventions and testing.
bray AGENTS.md
AGENTS.md instructions for forgesworn/bray, covering agents.md — nostr-bray, what this is, build & test, architecture and tool groups.