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/cdeust/cortex/whygit clone --depth 1 https://github.com/cdeust/CortexWhat 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 | $0.00027 | $0.00454 |
| Opus 5 | $0.00014 | $0.00227 |
| Sonnet 5 | $0.00005 | $0.00091 |
| Haiku 4.5 | $0.00003 | $0.00045 |
Grade A, and why
why 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.
What it actually says
Deterministic entry point for the blame path (decision 4255039, correction 6 — natural-language routing to the tool is not guaranteed; this command is).
- Scan the conversation context for
⟦rcpt:N⟧markers. They sit in the header line of every memory injection block (session-start banner, auto-recall block, agent briefing, recall responses'receipt_idfield). - Exclude the marker that arrived with THIS prompt's own memory-injection block (the auto-recall block directly attached to the current user message). That receipt records what was injected for the current question — including it would blame the question's own context (self-pollution guard, correction 4).
- If the user is questioning a specific earlier answer, keep only markers that were already in context before that answer was produced.
- Call the
whytool with the collected ids:why(receipt_ids=[...]). - Present the evidence with the locked lexicon:
- Say the memories were present in context ("présence-en-contexte") at the recorded instants and ranks. Never say a memory caused the answer — the receipts are Pearl-rung-1 evidence of presence, nothing more.
- Order as returned (emitted_at DESC, receipt id DESC as deterministic tiebreak, then persisted rank) — recorded facts only.
- Flag rows where
superseded_by_idis set (already corrected) and rows withmemory_missing: true(hard-forgotten after injection; the receipt remains valid presence evidence).
- If the user identifies a wrong memory among the evidence, offer to store a correcting memory with
remember(the write path supersedes near-duplicates; the explicit atomic correction flow lands in tranche 4).
If no ⟦rcpt:N⟧ marker is visible in context, say so honestly: no receipt in scope means no presence evidence to resolve — do not guess or reconstruct ids.
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.
- yesterday First seen · 19 lines · 27 tokens per session scan A 2937a2719777
why is a command published in the GitHub repository cdeust/Cortex (71 stars, last pushed 3d ago), licensed MIT. It adds 27 tokens to every session and 454 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
brief
Command "brief" from csmar432/finai-research, covering brief — 生成研究简报, arguments, 描述, 工作方式 and 示例.
validate-dependencies
Validate all task dependencies for issues.
flow-nexus-neural
Train and deploy neural networks in distributed sandboxes.
flow-nexus-auth
Flow Nexus authentication and user management.
validate-prd
Validate an existing PRD against BMAD standards - comprehensive review for completeness, clarity, and quality.
editorial-review-structure
Structural editor that proposes cuts, reorganization, and simplification while preserving comprehension.