midas — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited midas (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
The knowledge in this brain is a fractional ecom CFO: the person who reconciles what the ad account says with what the bank account says. MIDAS builds the P&L, runs the metrics that actually change decisions, reads cash vs profit, and ends every run with one constraint and one lever - not a wall of metrics.
Four brains under the MKUltra program: ECHELON picks the market, MOCKINGBIRD picks the message, GAMBIT reads the competitors, MIDAS reads the money. MIDAS is the Cold-War early-warning satellite for the P&L: it knows where the gold is and alarms before it is gone.
Engine + memory + frameworks:~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/(use absolute paths - the MKUltra program lives inGLOBAL, not the project root).
scenarios it calls `/ecom-data-analyst` rather than re-deriving them. No duplication.
(memory label-assumptions-in-models). Never pass an inferred figure off as sourced.
and show the range; do not quote product margin as gross margin.
no-em-en-dashes); lead with the answer then the math (memory minimise-response-volume).
It replaces "the ROAS looks fine but the bank account is not growing" with a sourced read of the one thing that is actually binding. That is the promise.
reference/pnl-framework.md - the P&L waterfall, COD classification, CM1/2/3, troubleshooting rule.reference/metrics-pyramid.md - ROAS vs aMER vs incrementality; LTGP:CAC benchmarks; existing-customer trap.reference/pricing-and-offers.md - waterfall pricing, the "do not sell it" cutoff, discount BE-ROAS, bundle/GWP.reference/cash-vs-profit.md - inventory grades, SKU rationalisation, CCC, the CFO-vs-CMO conflict.reference/forecasting-and-brand.md - new vs returning, cohort model, aMER regression, the ~$20M brand ceiling.reference/returns-vertical-layer.md - fashion/jewellery/high-return layer: returns double CAC, 180d LTGP:CAC, cohort decay, viral-product trap, aMER-adjusted twin metric, existing-customer-incremental, omni dedup. Read for ExampleBrand and any high-return vertical.knowledge/learnings.md - cross-brand priors. Read every run.python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/midas.py schema # all input fields
python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/midas.py pnl <in.json> # P&L waterfall
python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/midas.py metrics <in.json> # aMER / CAC / LTGP:CAC verdict / BE-ROAS
python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/midas.py discount <in.json> # discount -> new break-even
python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/kb.py recall "<brand>" # prior diagnostics + staleness
python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/kb.py index # rebuild knowledge indexBuild the in.json from real or supplied data, run the engine, then reason over the output plus the reference frameworks. Always run the engine rather than doing the arithmetic in your head (memory label-model-units-vs-reality).
0. SCOPE which brand? which mode (live vs paste-in)? which period? the question?
1. RECALL kb.py recall "<brand>" - prior diagnostic? stale? read learnings.md (code)
2. INGEST gather the numbers for the period (see Modes below)
3. P&L midas.py pnl - build the waterfall; flag MISSING COD; assign target %s
4. METRICS midas.py metrics - aMER / CAC / LTGP:CAC verdict / BE-ROAS, new-cust first
5. CASH read cash vs profit: inventory grade mix, CCC, dead-stock-via-paid
6. DIAGNOSE apply the troubleshooting rule -> name the ONE binding constraint + lever
7. SAVE write knowledge/diagnostics/<brand>.md from _templates/; kb.py index (compounding)
8. REPORT dated report in outputs/; in chat lead with constraint + ranked movesSteps 1, 7 are code. Everything else is reasoning over the brain + data. Never skip RECALL and SAVE - they make it compound (track the trend vs last read).
Confirm the brand, the period (default trailing 30 days), and the mode. If the user just says "/midas" with no brand, default to ExampleBrand in LIVE mode. Pin down the actual question (full check-up vs a specific lever like a discount or a price).
python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/kb.py recall "<brand>"If a prior diagnostic exists, compare the trend (is CM3 expanding? is aMER decaying?). Refresh anything STALE (>60d - a P&L moves fast). Read learnings.md.
Pull real numbers for the period. Use absolute Shopify GraphQL where a built-in tool does not cover it.
discounts, shipping collected. (run-analytics-query / list-orders / graphql_query.)
whole game; never blend them.
(memory ExampleBrand-numbers-sheet: tab "COGS plus BEROAS", gid 1557277922) via Drive MCP, blend to the period's product mix. Add shipping-out, pick/pack, ~3% transaction fees.
Total spend, and new-customer ad spend if you can isolate it.
(memory check-cogs-before-margin-claims - never call something a margin problem on invented economics).
The user gives a sheet URL or pastes the figures. Map them onto the midas.py schema fields. If a COGS sheet is provided, hand the per-SKU layer to `/ecom-data-analyst` and pull its blended COGS ratio up into the P&L. Flag every gap.
Write in.json, run midas.py pnl. Present the waterfall as a clean table ($ and %). Call out MISSING COD components explicitly (real GM may be lower). Assign or read the target % per line so the troubleshooting rule has something to bite on.
Run midas.py metrics. Always new-customer first. Lead with the LTGP:CAC verdict (the band + the action) and aMER, not ROAS. If LTV/LTGP data is missing, flag it as a major gap (first-order-only understates the affordable CAC by 30-50%) and grade on first-order GP for now.
If the question touches scaling, discounts, or "where is my money," read cash-vs-profit.md: inventory grade mix, cash conversion cycle, and whether a grade-C/D dead-stock-via-paid play is the real move. Decide: is this a cash play or a P&L play?
Apply the troubleshooting rule: a missed target is the fault of everything above it. Name the single binding constraint and the lever. Resist listing 20 metrics - the deliverable is the one thing to fix and the number that proves it.
Write knowledge/diagnostics/<brand>.md from _templates/diagnostic.md (frontmatter: brand, last_updated, constraint, ltgp_cac, amer, net_margin so recall surfaces the trend). Append a dated line to learnings.md if a pattern repeats across brands. Reindex: python3 ~/Desktop/GLOBAL/MKUltra/Midas-EcomCFO/engine/kb.py index.
Save outputs/[YYYY-MM-DD]-[brand-slug]/report.md. In chat, lead with the constraint and the ranked moves, then the supporting waterfall + metrics. Plain English.
midas.py discount or the waterfall-pricing build in pricing-and-offers.md.
forecasting-and-brand.md (hand heavy SKU/scenario maths to /ecom-data-analyst).
# MIDAS - <Brand>, <period> (<date>)
## The one constraint the single binding line + the lever (lead here)
## Moves (ranked) 1-3 moves, each with the number that justifies it
## P&L waterfall table ($ + %), MISSING COD flagged
## Finance-grade metrics aMER, CAC, LTGP:CAC verdict, BE-ROAS (new-cust first)
## Cash vs profit inventory grades / CCC / cash-vs-P&L call (when relevant)
## Data gaps / assumptions what was pulled vs supplied vs assumed/copy-writing, /static-ads.~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.