convergence-guard

convergence-guard is a skill for Claude Code, Codex from orenluxy/fable-method. It costs 74 tokens per session (590 once invoked), scanned A, original, MIT.

A stop-and-rethink guide for debugging and implementation sessions. It tracks repeated edits, failed fixes, and growing deviations from the original plan.

In plain words
What is it for?
Use it when tests keep failing, the same file is repeatedly changed, or a debugging loop needs a defined stopping point.
Why use it?
It helps prevent endless patching when the chosen approach is probably wrong and tokens are being spent without progress.

Skill for Claude CodeCodex

Part of the fable-method plugin — 8 skills, 2 agents, 3 hooks shipped together

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 skills/orenluxy/fable-method/convergence-guard
Any agent
npx skills add orenluxy/fable-method --skill convergence-guard
Clone the repo
git clone --depth 1 https://github.com/orenluxy/fable-method

Made for: Claude Code, Codex.

Or install fable-method, the plugin that ships this one along with the rest of its 8 skills, 2 agents, 3 hooks.

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 convergence-guard

README.md
[![agentmods](https://agentmods.dev/badge/skills/orenluxy/fable-method/convergence-guard.svg)](https://agentmods.dev/skills/orenluxy/fable-method/convergence-guard)
Your own site
<a href="https://agentmods.dev/skills/orenluxy/fable-method/convergence-guard"><img src="https://agentmods.dev/badge/skills/orenluxy/fable-method/convergence-guard.svg" alt="Measured on agentmods" height="20"></a>
Per session 74 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 590 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 $0.00074 $0.00590
Opus 5 $0.00037 $0.00295
Sonnet 5 $0.00015 $0.00118
Haiku 4.5 $0.00007 $0.00059

Measured 3d ago against content hash 72fcea4a84b4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

convergence-guard 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 3d 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.

skills/convergence-guard/SKILL.md · 38 lines

How it starts

The opening of the file, as written. The whole thing — 38 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Convergence Guard (stoploss)

Principle: the most expensive failure mode in agentic work is the fix-loop — repeatedly patching symptoms of a wrong plan. Tokens spent in a fix-loop are worse than wasted: they deepen commitment to a broken approach. This skill defines when to STOP.

External tool detection — check first

At session start (the PostToolUse hook does this automatically too): if a stoploss command exists on PATH or a stoploss.config.* file exists in the project, the user has a dedicated stoploss tool. Defer to it: run stoploss report (or its documented equivalent) when checking convergence, and treat its thresholds as authoritative over the defaults below.

Native thresholds (when no external tool is present)

Track mentally and via IMPLEMENTATION_NOTES.md:

  1. Same-file churn: the same file edited 5+ times in one session without its tests turning green → WARNING: state it explicitly and re-examine the approach before the next edit. 8+ edits → STOP.
  2. Three-strike rule: 3 consecutive fix attempts for the same failing test/behavior, each failing → STOP. Attempt #4 with the same strategy is forbidden.
  3. Deviation pressure: 3+ entries under Deviations in IMPLEMENTATION_NOTES.md touching the same subsystem → STOP.
  4. Scope creep: the session has touched 2x more files than the plan anticipated → pause and re-confirm scope with the user.

What STOP means

STOP is not "give up." It is a mandatory replan checkpoint:

  1. Freeze edits. No more code changes.
  2. Summarize for the user: what was attempted, what failed each time, what the failures have in common.
  3. State the hypothesis the fix-loop was implicitly betting on, and why the evidence now contradicts it.
  4. Propose 2 alternative approaches (different strategy, not different patch), with cost estimates.
  5. Wait for the user's call. Resuming the old strategy requires their explicit approval.

Anti-patterns

  • "One more try" past a threshold. Thresholds exist precisely because in-the-loop judgment is compromised in a loop.
  • Silently resetting the counter after a partial success. A test flipping green then red again counts as a strike, not a reset.
  • Treating STOP as failure. Detecting a wrong plan after 3 attempts instead of 15 is the system succeeding.

Read the full file on GitHub · 38 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. 3d ago First seen · 38 lines · 74 tokens per session scan A 72fcea4a84b4

Subscribe to this mod's changes

convergence-guard is a skill published in the GitHub repository orenluxy/fable-method (1 stars, last pushed 27d ago), licensed MIT. It adds 74 tokens to every session and 590 once invoked, about $0.0004 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 2026-08-31.

Related

Other skills, from other repositories

glossary-init

USE WHEN setting up a repo for AI-first work, when an agent hallucinates domain term meaning, or when user asks to bootstrap / extend the repo's domain glossary. Surfaces candidate terms by identifier frequency, asks the user for 1-sentence definitions, writes GLOSSARY.md at repo root. Idempotent — re-running adds new…

Filip-Podstavec/claude-leverage · 119 tokens

codex-pair-resume

Resume codex-pair for this project after a previous pause. Removes .codex-pair/state/paused and clears failure state. Review resumes on the next supported file edit; missing state is a no-op.

Lykhoyda/ask-llm · 51 tokens

generate-tests

Generate unit tests for a .NET service following the coverage-kit conventions. Use when the user asks to 'generate tests', 'backfill tests', 'characterize this service', or 'add tests for' a target after coverage-init has run. Operates in characterization mode (freeze current behavior for existing code) or spec mode…

livlign/claude-skills · 179 tokens

conventions-init

USE WHEN setting up a repo for AI-first work (after /init-repo), or when the context-surface hook should start feeding repo conventions to agents before edits. Drafts conventions.yml at repo root: per-kind casing (inferred from the code), a vague-name denylist seed, directory roles, and a hand-filled house-rules…

Filip-Podstavec/claude-leverage · 124 tokens

opik

This skill should be used when the user needs to add Opik tracing or integrations to their code, instrument an LLM application, or needs reference for Opik SDK usage (Python, TypeScript, REST API). Use for tasks like "add tracing", "instrument my code", "use trackopenai", "add OpikTracer", "what span types are…

comet-ml/opik-claude-code-plugin · 84 tokens

codex-review

Get a second opinion from OpenAI Codex on your current code changes. Analyzes staged/unstaged diffs and returns prioritized findings. Use when user asks to "review with Codex", "Codex code review", or "ask Codex to check my code".

Lykhoyda/ask-llm · 60 tokens