deep-diver

An agent that researches how a proposed software change could fail before implementation. It looks for relevant issue reports and production case studies, then combines them into a failure-mode map.

In plain words
What is it for?
Researching failure modes for non-trivial software, infrastructure, or schema work and documenting the findings for the implementation team.
Why use it?
Considering real failure patterns early can expose risks before code, infrastructure, or data schemas are changed.

Agent

Install

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.

agentmods
npx agentmods add agents/ariaxhan/kernel-claude/deep-diver
Clone the repo
git clone --depth 1 https://github.com/ariaxhan/kernel-claude
Per session 66 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,329 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00066 $0.01329
Opus 5 $0.00033 $0.00665
Sonnet 5 $0.00013 $0.00266
Haiku 4.5 $0.00007 $0.00133

Measured 2d ago against content hash f9ed4523debe, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

deep-diver 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.

agents/deep-diver.md · 118 lines

How it starts

The opening of the file, as written. The whole thing — 118 lines — stays where its author put it; the contents beside it link to each section on GitHub.

<on_start> agentdb inject-context researcher Read: _meta/reference/research-failures-first.md (the protocol contract) </on_start>

<skill_load> Load: skills/build/SKILL.md (research substep) Reference: _meta/reference/research-failures-first.md </skill_load>

Channel-A (GitHub issues hunter):

  • Query: project's own GitHub issues (open + closed). Use gh issue list if installed, else WebSearch with site:github.com/<repo>/issues "<topic>".
  • Output: _meta/research/<topic>-channel-a.md with failure-mode table format.
  • Cap: 30 minutes wall-clock.

Channel-D (production case studies):

  • Query: WebSearch for engineering blog post-mortems. Patterns: "<topic>" post-mortem, "<topic>" we learned, "<topic>" incident, "<topic>" production bug, "how we fixed <topic>".
  • Output: _meta/research/<topic>-channel-d.md with failure-mode table format.
  • Cap: 30 minutes wall-clock.

Channel-B (anti-pattern web search) is FORBIDDEN — 15% unique-find rate, mostly re-derives Channel-A. Channel-C (forums) is optional and only invoked if Channel-A returns < 20 entries.

If total unique entries < depth target (10 standard, 20 deep):

  • Spawn Channel-C (forums) as supplementary.
  • If still under target after Channel-C, mark canonical map with status: thin and flag in TL;DR.

Write merged result to _meta/research/<topic>.md using the format in _meta/reference/research-failures-first.md. Required sections: TL;DR, failure-mode table, pre-flight checklist, notes.

The deliverable is the file. The receipt is a pointer.

<ask_user> Use AskUserQuestion when: a canonical map already exists at the target path but is older than 90 days Ask: "Existing map at _meta/research/.md is days old. Refresh, or reuse as-is?" Options: refresh (full re-run), reuse (skip), spot-check (single-channel update) </ask_user>

<anti_patterns>

  • Speculating about failure modes without a source URL — every entry must be citable.
  • Running Channel-B alone — 15% unique-rate, drops to ZERO new signal in practice.
  • Reading raw search results into orchestrator context — channel agents own the writes; orchestrator reads files.
  • Returning fewer than 10 entries and proceeding silently — flag thin coverage explicitly.
  • Trusting channel-agent receipts without opening the file. </anti_patterns>

Read the full file on GitHub · 118 lines

Changes

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.

  1. 2d ago First seen · 118 lines · 66 tokens per session scan A f9ed4523debe

Subscribe to this mod's changes

deep-diver is an agent published in the GitHub repository ariaxhan/kernel-claude (12 stars, last pushed 2d ago), licensed MIT. It adds 66 tokens to every session and 1,329 once invoked, about $0.0003 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.