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 commands/flyfission/nuclear-grade-context-engineering/ng-learngit clone --depth 1 https://github.com/FlyFission/nuclear-grade-context-engineeringWhat 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 | $0.00000 | $0.00595 |
| Opus 5 | $0.00000 | $0.00298 |
| Sonnet 5 | $0.00000 | $0.00119 |
| Haiku 4.5 | $0.00000 | $0.00060 |
Grade A, and why
ng-learn 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 2d 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 — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ng-learn
Portable command prompt generated from skills/learning-from-experience/SKILL.md. Edit the skill, then run python tools/ng.py gen-commands; do not edit this file by hand.
Turns incidents, near misses, bad handoffs, review surprises, escaped bugs, and signals from real use into lasting fixes to your safeguards. Use after something went wrong or nearly did and a future safeguard should change. Do not use during a live incident, which comes first, or to blame someone.
Use when
- A bad handoff, a wrong-file edit, a made-up claim, an agent going past its allowed tools, a bug that escaped to users, or a surprise in review happened.
- Users or operators misread a release, a public claim, a runbook, a template, or an approved version.
- A past change record, skill, command, test, checker, monitor, or template failed to steer behavior the way it should have.
- A change to the rules or sources produced new text but no lasting change to a control.
Do not use when
- The event has no lesson that would repeat, and no control could reasonably change.
- You must contain a live incident first; analyze it after.
- The request is to blame someone rather than improve a control.
Inputs
- The event, near miss, review surprise, operating signal, or user feedback.
- The change record, approved version, file, skill, command, test, checker, monitor, or doc it affected.
- The evidence, the impact, the quick fix you already made, and the chance it happens again.
Prompt text
Create a Nuclear-grade OPEX record (lessons from real operation).
Inputs:
- event or near miss:
- affected packet / baseline / artifact:
- evidence:
- impact:
- immediate correction:
- weak or missing control:
- candidate durable update:
- owner:
- due date or trigger:
Produce a no-blame OPEX record (no-blame covers honest error, not a willful violation like a knowingly bypassed gate, disabled control, or fabricated result — surface those as findings, never file them as mistakes). Each finding must either change a lasting control or be closed with a clear reason why not.
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.
- 2d ago First seen · 54 lines · 0 tokens per session scan A cee41bd905fd
ng-learn is a command published in the GitHub repository FlyFission/nuclear-grade-context-engineering (33 stars, last pushed 24d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 595 tokens. 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 commands, from other repositories
coder-eval-implement-plan
Implement an approved codereval plan phase by phase with risk-scaled per-phase review, then a final code review.
capture-feedback
Quick feedback capture with structured signals.
auto-run
PitWay: Manage auto-run authorization for automatic task continuation.
task-integrate
PitWay: Apply a dispatched task's worktree commit to the main tree.
milestone-list
PitWay: List every milestone with a concise one-line status.
fix
Universal debugging and fix application with semantic code analysis.