Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/nguyenthienthanh/aura-frognpx agentmods add skills/nguyenthienthanh/aura-frog/failure-classifierWrote 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.
[](https://agentmods.dev/skills/nguyenthienthanh/aura-frog/failure-classifier)<a href="https://agentmods.dev/skills/nguyenthienthanh/aura-frog/failure-classifier"><img src="https://agentmods.dev/badge/skills/nguyenthienthanh/aura-frog/failure-classifier.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00052 | $0.01205 |
| Opus 5 | $0.00026 | $0.00602 |
| Sonnet 5 | $0.00010 | $0.00241 |
| Haiku 4.5 | $0.00005 | $0.00120 |
Grade A, and why
failure-classifier 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 6d 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 — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AI-consumed reference. Optimized for Claude to read during execution. Human-readable explanation: see docs/architecture/HIERARCHICAL_PLANNING.md or docs/getting-started/ depending on topic.
Failure Classifier
STATUS — v3.7.0-alpha.2 (Milestone B).
Classifies T4 task failures so master-planner can decide: retry, self-heal, replan, or escalate.
Failure classes
| Class | Name | Trigger | Recommended action |
|---|---|---|---|
| F1 | transient | Network timeout, file lock, ECONNRESET, rate limit, flaky CI | Retry with backoff (max 3) |
| F2 | local-logic | Test fails, type error, lint error, assertion mismatch | Self-heal-propose (Milestone E) or single retry |
| F3 | local-design | Story acceptance unreachable as designed; signals: same test fails after 2 fixes, or output drift > 50% from plan | Replan story (re-decompose T3) |
| F4 | story-level | Sibling tasks broken, dependency contract violated, deviation_score ≥ 0.7 | Replan feature (re-decompose T2) |
| F5 | architectural | Cross-feature impact, mission-level constraint violated | Freeze + escalate to user |
F6 — conflict-induced (added in v3.7.0-beta.2): when cause: conflict is set on the failure signal OR conflict_id is non-null, classifier returns class F6 with confidence 0.95 and routes to conflict-arbiter agent (NOT replanner). The arbiter applies the decision table from rules/workflow/conflict-arbitration-policy.md.
Classifier inputs
- Failure event from post-execute-update-node hook:
{tool, exit_code, stderr_tail, files_touched} - Active task node (status, failed_attempts, replan_budget_remaining)
- Last 10 history.jsonl events for the task
- Failed test output (when available)
Decision rules (deterministic, evaluated top-down — first match wins)
1. exit_code == 0 → not a failure; abort classification
2. failure.cause == "conflict" OR failure.conflict_id is non-null → F6
3. stderr matches /ECONNRESET|ETIMEDOUT|EAI_AGAIN|429|503|rate limit/i → F1
4. failed_attempts >= 3 AND same stderr_hash as last 2 attempts → F3
5. files_touched intersects sibling task's files_touched (from history) → F4
6. failed_attempts == 1 AND stderr matches /TypeError|SyntaxError|test failed|assertion/ → F2
7. deviation_score(node) >= 0.7 → F4
8. ancestor_T1.intent contradicted by failure evidence → F5
9. default → F2
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.
- 6d ago First seen · 98 lines · 52 tokens per session scan A a098bb9096e3
failure-classifier is a skill published in the GitHub repository nguyenthienthanh/aura-frog (24 stars, last pushed 2d ago), licensed MIT. It adds 52 tokens to every session and 1,205 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.
Other skills, from other repositories
biome
Biome formatting and linting standards. Use when configuring Biome (biome.json) for fast formatting, linting, and sorting imports. NOT for Prettier or ESLint configuration. Do NOT trigger for generic formatting/linting tasks unless Biome is explicitly requested.
adversarial-performance
Multi-pass adversarial performance audit for entire repositories. Combines structured profiling (Agent A) with adversarial stress-testing critique (Agent B) through iterative passes. Optimize repo/backend performance, hot-paths, build size, and server throughput. Do NOT use for frontend page load metrics or Lighthouse…
autonomous-dev
Harness for autonomous software development. Use when fixing or remediating known issues (including security vulnerabilities). Enforces lifecycle through alignment gates (PROJECT.md), adversarial generator/evaluator agents, and autonomous orchestration of project issues. NOT for setting up standalone CI/CD pipelines.
debugger-coordinator
Use when debugging complex issues involving multiple modalities (frontend + backend, API + UI, terminal + browser). Coordinates Hermes Agent's built-in debugger, browser, terminal, and vision tools for systematic root-cause analysis.
gsd-debug
Systematic debugging with persistent state across context resets.
gsd-ns-review
Route to the appropriate quality / review skill based on the user's intent. gsd-code-review-fix was absorbed by gsd-code-review --fix in #2790.