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/ashkaan/contextium/spec-auditgit clone --depth 1 https://github.com/Ashkaan/contextiumWhat 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 | $0.00015 | $0.00261 |
| Opus 5 | $0.00008 | $0.00130 |
| Sonnet 5 | $0.00003 | $0.00052 |
| Haiku 4.5 | $0.00002 | $0.00026 |
Grade A, and why
spec-audit 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.
What it actually says
Review the SPEC you just wrote, adversarially, before it becomes code. A design flaw caught here costs a paragraph; the same flaw caught after the code exists costs the implementation.
- Read the SPEC against what was actually asked. Does it describe that thing, or a fancier interpretation of it? A SPEC can be internally excellent and solve the wrong problem.
- Attack the design: missing edge cases (0 / 1 / empty / max / error), downstream callers it doesn't mention, sibling code it leaves inconsistent, failure modes it never names, a shape heavier than the ask, and a "done" section that isn't an actual runnable command with an actual expected output.
- For each finding, either fix the SPEC now or write down why the finding is wrong. Those are the only two options — "track it for later" is how a known flaw ships.
In Claude Code this runs on a different model from the one that wrote the SPEC, when one is installed. In other tools, open a clean chat — ideally in a different assistant — load only the ask and the SPEC, and run this review there. The author is the worst reviewer of their own design.
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 · 12 lines · 15 tokens per session scan A 2401dbb2d03f
spec-audit is a command published in the GitHub repository Ashkaan/contextium (32 stars, last pushed 5d ago), licensed MIT. It adds 15 tokens to every session and 261 once invoked, about $0.0001 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 commands, from other repositories
test-tools
Test MCP NixOS Tools (project).
release
Review and publish the Release Please release PR, then verify every registry.
design-qa
Agent-driven design QA — walk the running app, judge design/UX, and drop a Comment-Mode comment on every issue (visible to the human as pins). Then read the session back and fix everything in bulk. Optional focus argument steers the pass.
best-of
Spawn N parallel code-task sub-agents in worktrees and apply the winner's diff (code/file-change tasks only).
checkpoint
Write a new link in the continuum chain summarizing decisions/changes since the last link.
dream
Roll up the last N links into a single phase-digest link; originals move to archived/ (never deleted).