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 skills/kastalien-research/thoughtbox/thoughtbox-debugnpx skills add Kastalien-Research/thoughtbox --skill thoughtbox-debuggit clone --depth 1 https://github.com/Kastalien-Research/thoughtboxWhat 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 | $0.00080 | $0.01056 |
| Opus 5 | $0.00040 | $0.00528 |
| Sonnet 5 | $0.00016 | $0.00211 |
| Haiku 4.5 | $0.00008 | $0.00106 |
Grade A, and why
thoughtbox:debug scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Thoughtbox Debug
Guided debugging workflow built on the Ulysses surprise-gated protocol. Prevents reactive debugging spirals by forcing structured hypotheses after repeated surprises.
Core Mechanic: The S-Register
The S-register (surprise counter) is the discipline mechanism. Each unexpected outcome increments S. At S=2, stop and form a falsifiable hypothesis before continuing. This breaks the "try something, doesn't work, try something else" spiral.
Phase 1: Initialize
Start a debugging session. Define the problem and any constraints.
async () => {
return await tb.ulysses({
operation: "init",
problem: "$ARGUMENTS",
constraints: [
// Hard limits on what you can change
]
});
}
This starts a session with S=0. State the problem precisely — vague problems produce vague debugging.
Phase 2: Plan-Act-Assess Loop
Repeat this cycle for each investigation step.
2a. Plan with a pre-committed recovery action
Before investigating, declare what you will do AND what you will do if it fails. The recovery action prevents post-hoc rationalization.
async () => {
return await tb.ulysses({
operation: "plan",
primary: "Check CI environment variables vs local .env",
recovery: "If env vars match, check node version differences"
});
}
2b. Execute
Run the primary action using whatever tools are needed (Read, Grep, Bash, etc.). Gather evidence. Do not interpret yet.
2c. Assess the outcome
Report whether the result matched your expectation.
async () => {
return await tb.ulysses({
operation: "outcome",
assessment: "unexpected", // or "expected"
severity: "minor", // or "major"
details: "Env vars are identical — rules out config differences"
});
}
"unexpected"increments S."expected"leaves S unchanged.- Be honest. Calling a surprise "expected" defeats the protocol.
Phase 3: Forced Reflection (S=2)
When S hits 2, the protocol blocks further plan operations until you reflect. This is the mechanism that prevents spiraling.
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.
- 3d ago First seen · 150 lines · 80 tokens per session scan A 45742f29235c
thoughtbox:debug is a skill published in the GitHub repository Kastalien-Research/thoughtbox (64 stars, last pushed 1mo ago), licensed MIT. It adds 80 tokens to every session and 1,056 once invoked, about $0.0004 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 2026-08-30.
Other skills, from other repositories
reasoning
Use BEFORE answering analytical, diagnostic, planning, or multi-step reasoning questions. Trigger phrases include "should I X or Y", "why is X happening", "what's the best approach", "what are the tradeoffs", "help me think through", "diagnose", "root cause", "plan/design X", "what are the implications of", "compare…
xcodebuildmcp-structured-output-review
Use when reviewing XcodeBuildMCP structured output schema changes, schema versioning, manifest outputSchema metadata, and JSON fixture compatibility.
opik-diagnose
Surface the Opik traces worth a developer's attention, ranked by signal — errors, failed tool calls, latency, regressions, and low online-eval scores — plus Diagnostics issues. Reads live/production traces via the SDK (searchtraces and agentinsights) and works with no MCP; uses the MCP issue entity when connected.…
cortex-automate
Set up automation — prospective memory triggers, neuro-symbolic rules, and CLAUDE.md sync. Use when the user says 'remind me when', 'trigger when', 'create a rule', 'auto-remember', 'sync to CLAUDE.md', 'push insights', 'set up trigger', 'when I open this file', 'when this keyword appears', or when you want to…
tabnexus-mcp-evals
Generate, validate, and run isolated Codex-to-TabNexus MCP evaluations with a curated 600-query dataset, executable gold tool labels, safety checks, and best-of-three stability scoring. Use when testing TabNexus MCP tool coverage, Agent behavior, regression quality, destructive-action safety, prompt changes, or a…
compare
Structured comparison of 2+ alternatives with consistent criteria and decision matrix.