A guide for adding a self-healing workflow to a new engine. A self-healing workflow detects a broken result, rebuilds it from trusted source data, records the change, and checks the result again.

Use it when building a new engine that needs independent checks, trusted data reconstruction, approval gates, and realistic fault demonstrations.

It provides a repeatable way to repair drift without guessing or copying the wrong approval rules from another system.

How sophonfinance-wq/finance-automation-portfolio configures Claude Co

Reuse

Borrowing it

Nothing to install: this file belongs to sophonfinance-wq/finance-automation-portfolio. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file

curl -O https://raw.githubusercontent.com/sophonfinance-wq/finance-automation-portfolio/main/.claude/skills/build-loop/SKILL.md

Clone the repo

git clone --depth 1 https://github.com/sophonfinance-wq/finance-automation-portfolio

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.

README.md

[![agentmods](https://agentmods.dev/badge/skills/sophonfinance-wq/finance-automation-portfolio/build-loop/github.svg)](https://agentmods.dev/skills/sophonfinance-wq/finance-automation-portfolio/build-loop)

Your own site

<a href="https://agentmods.dev/skills/sophonfinance-wq/finance-automation-portfolio/build-loop"><img src="https://agentmods.dev/badge/skills/sophonfinance-wq/finance-automation-portfolio/build-loop/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

Your own site · 80×15

<a href="https://agentmods.dev/skills/sophonfinance-wq/finance-automation-portfolio/build-loop"><img src="https://agentmods.dev/badge/skills/sophonfinance-wq/finance-automation-portfolio/build-loop.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 47 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.

When invoked 580 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.

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.

Model	Per session	Once invoked
Fable 5.1	$0.00047	$0.00580
Opus 5.5	$0.00019	$0.00232
Sonnet 5.5	$0.00009	$0.00116
Haiku 4.5	$0.00005	$0.00058

Measured 2mo ago against content hash e4006cad222a, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

build-loop 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 2mo 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.

.claude/skills/build-loop/SKILL.md · 19 lines

How it starts

The opening of the file, as written. The whole thing — 19 lines — stays where its author put it; the contents beside it link to each section on GitHub.

How to Build a Loop (house methodology)

Reference implementations: surplus_engine/loop.py (human-gated), close_engine/loop.py (autonomous+quarantine), recon_engine/loop.py (materiality), brain_engine/loop.py (verbatim-or-refuse), triangulate/loop.py (arithmetic-vs-judgment).

Recipe

Scout the engine first: find the SENSOR (an existing independent check that yields named breaks) and the AUTHORITY (a deterministic re-derivation from the seeded source of record). If no sensor exists (brain case), build one that compares stored state to source.
Pick the gate policy from the domain's discipline — never copy another engine's. Ask: what is this system entitled to decide on its own? What must it refuse to touch (quarantine/halt/escalate)?
Loop core: while breaks remain and turns < budget: pick the EARLIEST failing scope (period/category/account/meeting/cell), resync it to authority, book every field change as an Adjustment, re-run the sensor. Frozen dataclasses -> dataclasses.replace; mutable -> setattr.
Demo faults: 3-5 realistic drift injectors, each with a root-cause story, targeting scopes chosen dynamically from the baseline (don't hardcode IDs). Include at least one fault the loop must REFUSE to fix.
Deterministic journal (frozen dataclasses): faults, initial findings, turns (before/after counts, adjustments), quarantined/held items, verdict. render_markdown(journal) must be byte-stable.
CLI: --demo, --budget, --out; verdict doubles as exit code (match the engine's existing convention — e.g. brain refusal=3).
Tests (~10-12, mirror tests/test_loop.py files): drift trips the sensor; loop converges; NEVER-INVENTS (settled state equals clean authority run exactly); the refusal boundary holds; clean input = zero turns; budget exhaustion fails; exit codes; determinism (two runs render identical markdown); CLI smoke with tmp_path.
README section: what/why, the five-stage loop line, verdict table with exit codes, --demo command + real output excerpt.
Run the package suite AND the full portfolio (python -m pytest per package, like CI) before pushing.

Read the full file on GitHub · 19 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.

2mo ago First seen · 19 lines · 47 tokens per session scan A e4006cad222a

Subscribe to this mod's changes

build-loop is a skill published in the GitHub repository sophonfinance-wq/finance-automation-portfolio (16 stars, last pushed 27d ago), licensed MIT. It adds 47 tokens to every session and 580 once invoked, about $0.0002 per session on Opus 5.5. 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.

Debugging Python accounting ai automation finance-automation month-end-close

build-loop next to 6 skills from other repositories
Mod	Type	Stars	Last push	Licence	Scan	Tokens	Works with
build-loopsophonfinance-wq/finance-automation-portfolio	Skill	16	27d ago	MIT	A	47	Claude Code
developing-with-streamlitstreamlit/streamlit	Skill	46k	today	Apache-2.0	A	146	Claude Code, Codex
python-code-qualitymicrosoft/agent-framework	Skill	15k	today	MIT	A	40	Claude Code, Codex
eveComposioHQ/composio	Skill	31k	today	MIT	A	47	Claude Code, Codex
plugin-architecture-patternsxberg-io/xberg	Skill	9.6k	today	MIT	A	15	Claude Code
test-corpusxberg-io/xberg	Skill	9.6k	today	MIT	A	15	Claude Code, Codex
typescript-testingComposioHQ/composio	Skill	31k	today	MIT	A	92	Claude Code, Codex

finance-automation-portfolio: Skill for Claude Code