silent-failure-hunter

silent-failure-hunter is an agent for Claude Code from Kastalien-Research/thoughtbox. It costs 64 tokens per session (2,297 once invoked), scanned B, original, MIT.

A code-review agent that looks for failures which produce no visible error, especially in hooks, state-file pipelines, scheduled tasks, and error handlers.

In plain words
What is it for?
It is for auditing code paths, hooks, scripts, and error handling for silent failures, misleading success results, and plausible but wrong output.
Why use it?
It targets bugs that appear to work but quietly create incorrect data or skip updates. This helps find problems that ordinary crash-based debugging can miss.

Agent for Claude Code

Written for Claude Code: installed under .claude/. Also seen: model in frontmatter; reads .claude/ paths.

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/kastalien-research/thoughtbox/silent-failure-hunter
Clone the repo
git clone --depth 1 https://github.com/Kastalien-Research/thoughtbox

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for silent-failure-hunter

README.md
[![agentmods](https://agentmods.dev/badge/agents/kastalien-research/thoughtbox/silent-failure-hunter.svg)](https://agentmods.dev/agents/kastalien-research/thoughtbox/silent-failure-hunter)
Your own site
<a href="https://agentmods.dev/agents/kastalien-research/thoughtbox/silent-failure-hunter"><img src="https://agentmods.dev/badge/agents/kastalien-research/thoughtbox/silent-failure-hunter.svg" alt="Measured on agentmods" height="20"></a>
Per session 64 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,297 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. 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 5.1 $0.00064 $0.02297
Opus 5 $0.00032 $0.01149
Sonnet 5 $0.00013 $0.00459
Haiku 4.5 $0.00006 $0.00230

Measured 2d ago against content hash eb113abbe825, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade B, and why

silent-failure-hunter scanned grade B with 1 finding 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.

Reads agent configuration directoriesmediumAgent snooping

.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.

grep -rn '2>/dev/null' --include="*.sh" .claude/hooks/ specs/continual-improvement/hooks/
.claude/agents/silent-failure-hunter.md · 232 lines

How it starts

The opening of the file, as written. The whole thing — 232 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Silent Failure Hunter. You find code that breaks without telling anyone.

The most expensive bugs in this project have been silent failures — code that runs successfully but produces wrong results. eval_collector.sh returning memory_usefulness: 5 on every run. The fitness tracker skipping updates because a file didn't exist. Hook scripts that exit 0 without doing their job.

Your specialization: you don't look for crashes. You look for code that looks like it's working.

Playbook Loading

Load the shared adversarial playbook, filtered for your specialty:

sqlite3 research-workflows/workflows.db "
  SELECT attack_pattern, target_type, hit_rate, avg_severity
  FROM adversarial_findings
  WHERE agent = 'silent-failure-hunter'
    AND hit_rate > 0.2
  ORDER BY hit_rate * avg_severity DESC
  LIMIT 15;
" 2>/dev/null || echo "No playbook yet"

Also load the devil's advocate findings to avoid duplicating work:

sqlite3 research-workflows/workflows.db "
  SELECT target_file, finding FROM adversarial_findings
  WHERE agent = 'devils-advocate'
    AND found_at > datetime('now', '-7 days')
  LIMIT 20;
" 2>/dev/null || echo "No recent DA findings"

Hunting Protocol

Phase 1: Map the Data Pipelines (30% of turns)

Build a map of every data pipeline in the target scope:

[Producer] → writes → [State File] → reads ← [Consumer]

For this project, the critical pipelines are:

Producer State File Consumer Risk
fitness_tracker.sh .dgm/fitness.json /dgm-evolve Pattern evolution uses wrong fitness
eval_collector.sh .eval/metrics/session-*.json regression-sentinel Regression detection uses wrong baselines
assumption-tracker.sh .assumptions/registry.jsonl assumption-auditor Stale assumptions not flagged
session_end_memory.sh .claude/state/memory-calibration.log eval_collector.sh Memory usefulness always returns default
controller-prime.sh controller/state/controller-state.json session_start.sh Session gets no priming context
Agent SDK scripts historical run-log directories cost-governor Cost tracking misses runs

Read the full file on GitHub · 232 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 · 232 lines · 64 tokens per session scan B eb113abbe825

Subscribe to this mod's changes

silent-failure-hunter is an agent published in the GitHub repository Kastalien-Research/thoughtbox (64 stars, last pushed 1mo ago), licensed MIT. It adds 64 tokens to every session and 2,297 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.

Related

Other agents, from other repositories

semble-search

Code search agent for exploring any codebase. Use for finding code by intent, locating implementations, understanding how something works, or discovering related code. Prefer over runshellcommand/readfile for any semantic or exploratory question.

MinishLab/semble · 46 tokens

challenger

Use when: before the lead reports a root-cause conclusion, a 'done/verified' claim, an irreversible action about to run (commit/deploy/rm/push), or a 2nd-time fix — in APEX or plain conversation; also every eLicit round and Verify gate. Do NOT use for: code correctness/lint/types/API usage (sniper's job), or as a veto…

fusengine/agents · 92 tokens

sniper

Use when: after ANY code modification (mandatory post-edit validation). Do NOT use for: new features, quick fixes already identified (use sniper-faster), read-only analysis.

fusengine/agents · 38 tokens

semble-search

Code search agent for exploring any codebase. Use for finding code by intent, locating implementations, understanding how something works, or discovering related code. Prefer over Grep/Glob/Read for any semantic or exploratory question.

MinishLab/semble · 47 tokens

semble-search

Code search agent for exploring any codebase. Use for finding code by intent, locating implementations, understanding how something works, or discovering related code. Prefer over Bash/Read for any semantic or exploratory question.

MinishLab/semble · 44 tokens

semble-search

Code search agent for exploring any codebase. Use for finding code by intent, locating implementations, understanding how something works, or discovering related code. Prefer over Bash/Read for any semantic or exploratory question.

MinishLab/semble · 44 tokens