sales-movers-scanner
Skill8 starsChanged 45 days agoMITView on GitHub
Scan the whole catalog for the SKUs that moved the most this period - up and down - and decompose each move into its cause: traffic, conversion, price, or buy-box. Ranked by the size of the swing in money, with a data-completeness guard on both windows so a lagging or partial week doesn't read as a fake collapse or a fake boom. Live from DataDoe, read-only. Use for "what changed", "biggest movers", "why did sales drop", "what's up this week", "sales down", "which products dropped", "what moved", or "week over week".
In this repository26 other files it gives its agents.
- amazon-asin-search-auditorSkill
- amazon-listing-optimizerSkill
- buy-box-loss-root-causeSkill
- create-amazon-reconciliation-dashboardSkill
- create-orders-managerSkill
- daily-account-health-checkSkill
- daily-performance-health-checkSkill
- fba-inventory-ledger-reportSkill
- fba-stock-restock-plannerSkill
- keyword-rank-sqp-trackerSkill
- net-profit-pl-analyzerSkill
- ppc-asin-acos-alertSkill
- ppc-bid-optimizer-applySkill
- ppc-campaign-acos-alertSkill
- ppc-negative-keyword-applierSkill
- ppc-profit-bid-applySkill
- ppc-wasted-spend-watchdogSkill
- promotions-managerSkill
- restock-priority-alertSkill
- return-refund-analyzerSkill
- suppressed-inactive-listings-checkSkill
- vendor-inventory-reportSkill
- vendor-performance-dashboardSkill
- vendor-pnl-dashboardSkill
- weekly-business-reviewSkill
- weekly-sales-briefingSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 34
After a price change, ad change, or a competitor move, to see the ripple per SKU.
- Distinctive: line 101
NOT report the move as real.
- Distinctive: line 114
The uniform-move tell: If *most* SKUs move by a *similar* amount in the *same* direction and *all* read as "traffic" (sessions up/down …
- Distinctive: line 373
Read-only (analysis). Fixes that follow (price, listing, ads) are separate write skills, each dryRun-gated.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Sales Movers Scanner
Scans the whole catalog for the SKUs that moved the most this period - up and down - and decomposes each move into its cause: traffic, conversion, price, or buy-box. Instead of "sales dropped," it tells you "sales dropped because sessions fell 40%" or "because you lost the buy box" - so you act on the reason, not the symptom. Ranked by the size of the swing in money, so you see what actually matters. Live from DataDoe, read-only.
When to use this
- Weekly review: "what changed across my catalog, and why?"
- Revenue is up or down overall and you need the SKUs and reasons behind it.
- Monday-morning triage - a short list of what needs attention, already diagnosed.
- After a price change, ad change, or a competitor move, to see the ripple per SKU.
- Trigger phrases: "what changed", "biggest movers", "why did sales drop", "what's up this week", "sales down", "which products dropped", "what moved", "week over week".
The framework. Rank the movers, decompose the cause
Sales for a SKU are roughly sessions x conversion x price. So a change in sales comes from one (or more) of three levers, and the buy-box sits behind conversion. Compare a recent window to the prior equal window, then for each big mover decompose:
- Size the move - rank SKUs by the absolute change in sales (money), not by percent (a 5% drop on a hero SKU beats a 90% drop on a trickle SKU). Look at gainers AND decliners.
- Decompose the driver for each mover:
- Traffic -
total_sessions(ortotal_page_views), per child ASIN per day onamazon_profit_by_sku_and_date. Down/up means a visibility, rank, ads, suppression, or seasonality change. Sales followed the traffic. - Conversion - there is no conversion-rate column on this table; compute it as
total_units_sold / total_sessions(unit session rate; n/a when sessions are 0 - a few ASINs sell with no recorded sessions) ortotal_orders / total_sessions(order rate) per ASIN per window, after aggregating per the per-ASIN rule in Configuration. Conversion down/up while traffic held -> listing/price/reviews/offer problem or win. - Price / AOV - sales per unit (
total_sales / total_units_sold) shifted -> a price change, promo, or mix shift, even if units held. - Buy-box -
avg_buybox_percentagedown is the classic hidden cause of a conversion drop (you still get traffic but can't convert it). Always check it on a decliner. - Availability - a mover whose
avg_buybox_percentage(or conversion) rises from ~0 is usually back in stock / newly buyable, not an organic win; label it that way. A computed conversion above 100% is a units-per-session quirk (multi-unit orders), not a literal rate - don't report it as-is.
- Traffic -
Name the dominant driver per ASIN with the log split, which adds up exactly (S = sales, cvr = units / sessions, price = sales / units; 0 = prior window, 1 = recent):
ln(S1/S0) = ln(sessions1/sessions0) + ln(cvr1/cvr0) + ln(price1/price0)The driver is the term with the largest absolute value in the direction of the move (the most negative term for a decliner, the most positive for a gainer). Relabel a conversion-led move as buy-box when the page-view-weighted buy box moved by at least 10 points in the direction of the move (fell for a decliner, rose for a gainer); a rise from ~0 is availability (back in stock / newly buyable), not a buy-box win. When sessions are 0 in either window the split is undefined: mark the driver n/a (no traffic data) rather than guessing. When units are 0 in either window, it is availability (went to or came back from zero - see below). Then route it: traffic -> rank/ads/suppression check; conversion -> listing/price/reviews (or, on a gainer, a listing/offer win to protect); buy-box -> pricing/competitor (or, on a gainer, the offer won back - hold the price and stock that won it); price -> confirm the change was intended.
- Roll up: is the catalog move concentrated in a few SKUs or broad? One suppressed hero SKU vs an across-the-board seasonal dip are very different stories.
Data-completeness guard (check BOTH windows before you trust any broad move). Sales run ~1 day behind (shipped-order sales, refreshed intraday), traffic runs behind by a lag you must measure (Amazon says up to 3 days; measured from 2-3 days (Nov 2026) to 8-13 days (Oct 2026) on a UK account, and per-ASIN traffic can trail the account total by a day) and is revised for up to 30, and both can have mid-window holes, so either window can be under-counted - a lagging recent window fakes a collapse, and an under-counted prior window fakes a boom (every SKU reads as a riser). Guard both:
- Normalize by effective days, not calendar days present. A window's effective
days = sum over its days of (ASINs reporting that day / median daily ASIN count),
counted from the per-day rows of the window exports (step 4) - once for sales (ASINs
with a row) and once for traffic (ASINs with
total_sessions > 0). Two windows are only comparable when their effective days are ~equal: if they differ by more than 0.3 days on either count (a missing or half-loaded day on either side), the delta is an artifact - re-pull/shift the windows or state that a window looks under-reported; do NOT report the move as real. - Traffic completes later than sales - and per ASIN later than per account.
total_sessions,total_page_viewsandavg_buybox_percentagecome from Amazon's Sales and Traffic report and may be delayed (Amazon says up to 3 days; measured from 2-3 days (Nov 2026) to 8-13 days (Oct 2026) on a UK account, with holes, and per-ASIN traffic can trail the account total by a day - measure it) and revised for up to 30, while sales are ~1 day behind. A day whose traffic is not published yet reads as 0 sessions, 0 page views and 0 buy box on every ASIN - the table cannot tell that from a real zero. Judge traffic completeness separately from sales and per ASIN (step 3): the days after the last traffic-complete day are traffic-provisional even when their sales are complete, and a traffic-driven "decline" confined to those days is the lag, not a move. Say so in the output whenever the recent window includes them. - The uniform-move tell. If most SKUs move by a similar amount in the same direction and all read as "traffic" (sessions up/down ~uniformly), that is almost always an incomplete window on one side, not a real catalog swing. A real move is concentrated in specific SKUs with mixed drivers (traffic here, buy-box there), not a flat uniform shift everywhere.
Configuration
- MCP base:
https://mcp.datadoe.com/mcp/v1 - Data sources (resolve by table name with
exports_sources_get). Both are premium export (5 AI Tokens instead of 2; always-on default dataset, never disabled) - do not print the token cost in the output - and both haverequiresDatePeriod: true, so top-levelfrom/toare mandatory on every export (adatefilter does not replace them):amazon_profit_by_date- "Profit by Date", account level, one row per date and currency:total_sales,total_units_sold,total_orders,total_sessions. Used only to calibrate completeness and find the window end (step 3).amazon_profit_by_sku_and_date- "Profit by SKU & Date", one row per SKU, child ASIN and date: the outcome (total_sales,total_units_sold,total_orders) and the funnel (total_sessions,total_page_views,avg_buybox_percentage) side by side, pluschild_asin,sku,product_name. It replaces the deprecated Sales & Traffic by ASIN & Date table this skill used to build on. Per-ASIN aggregation rule:total_sessions,total_page_viewsandavg_buybox_percentageare "for the child ASIN on this date" and are repeated on every SKU row of that ASIN - grouping bychild_asinand summing them double-counts every ASIN with more than one SKU. Group by[date, child_asin, ...]and takemax()for those three;sum()is only right for sales, units and orders. Conversion is not a column: computetotal_units_sold / total_sessionsin code.total_salesis shipped-order sales, VAT-inclusive where Amazon reports prices with tax (sales_taxholds the tax part if an ex-VAT view is wanted).
- Windows: compare a recent window to the prior equal window (e.g. last 7 days vs the 7 before). Sales run ~1 day behind (the newest day is still filling as orders ship), traffic by a lag that varies (2-3 to 8-13 days measured on one UK account, per ASIN a day more), and holes happen - never assume a fixed offset. Detect the last complete day dynamically (see workflow step 3) - it is the end of the latest contiguous run of complete days, not an isolated complete day sitting after a hole - and end both windows there, so you compare two equally-complete windows, not a full week against a half-reported one.
- Currency/marketplace: read
marketplace_country_codeandcurrency; keep each marketplace in its own currency and scan them separately - never sum sales across currencies.
Step-by-step workflow (MCP-native)
sellers_and_vendors_list-> pick the seller.exports_sources_get-> resolve thesourceIdofamazon_profit_by_date(query "profit by date") and ofamazon_profit_by_sku_and_date(query "profit by sku"), confirming each hit ontableName. Both are in the always-on default dataset, so there is no deprecation warning to carry into the output header.- Calibrate completeness first - don't assume a lag.
exports_createonce onamazon_profit_by_dateover a wide span (~45 days - the traffic lag alone can eat two weeks) withgroupBy: ["date", "currency"](rows are per date and currency) and aggregationssum(total_sales)assales,sum(total_units_sold)asunitsandmax(total_sessions)assessions(account-level rows are already one per date and currency, somaxjust carries the day's value through),orderByColumn: date, JSON,limit: 100. One call, one row per day per currency (~23-30 rows each). This table is account level - there is no per-ASIN row count here; the per-ASIN traffic count comes from the one probe export below, and the per-day ASIN counts for effective days come free from the per-day window exports in step 4. From the per-day values:- Median daily sales, units and sessions for the account over the span (leave the newest 3 days out of the medians), and a full-day threshold = ~60% of each median - relative and per-account, never a fixed amount.
- Sales-complete day = a
datewhosesalesandunitsare at or above their threshold. Expect the newest day (sometimes two) to miss it - that is the normal ~1-day shipped-sales lag, not an outage; a day under the threshold inside the span is a hole. Mandatory cross-check: addsum(unshipped_sales); a day where it exceeds ~10% oftotal_sales + unshipped_salesis still filling and is NOT complete even if it passes the 60% test (observed: a day at 74% of median sales with 18.5% unshipped flipped the week-over-week sign). Keep the per-day unshipped values - step 5 sums them per window. - Traffic-complete day - judged separately from sales, because traffic lags
more than sales (measured from 2-3 to 8-13 days on one UK account, holes happen) and
is revised for up to 30; measure the tail, never assume it. Two tests, both needed:
- Account level (necessary, not enough): the day's
sessionsare at or above their threshold. - Per ASIN (decides it): per-ASIN traffic can trail the account total by a day or
more (observed: 5,433 account sessions on a day where 0 child ASINs had
sessions, against a median of ~1,390 per day - every product then read as a
traffic drop and every buy box fell ~15 points). Run one probe export on
amazon_profit_by_sku_and_dateover the same span:filterstotal_sessions > 0andmarketplace_country_code= the scanned marketplace (the column holds the ISO code -GBfor the UK, notUK; aninfilter with the valueGB,UKworks when unsure). Without the marketplace filter, a seller whose marketplaces share most child ASINs passes a day on another marketplace's traffic, and every product of the scanned one reads as a traffic drop again.groupBy: ["date"], aggregationcountDistinctonchild_asinasasins_with_sessions,orderByColumn: date, JSON,limit: 100(one row per day, ~45 rows). A date with no ASIN traffic at all returns no row - count it as 0. The day passes when its count is at least ~60% of the median daily count (the same relative rule as sales; leave the newest 3 days out of the median).
- Account level (necessary, not enough): the day's
- Last complete day (window end) = the latest date D such that D and the 6 days before it are all sales-complete - i.e. the end of the latest contiguous complete run. An isolated complete day after a hole is not an anchor (e.g. complete 08-30..09-20, hole 09-21..09-24, one day 09-25 -> window end 09-20, not 09-25). If no 7-day contiguous run exists in the span, widen the span once; failing that, anchor to the latest complete day and flag the scan provisional.
- Traffic window end = the end of the latest contiguous run of days that pass both traffic tests. When it is earlier than the sales window end, either end both windows at the traffic end (one clean comparison; the default whenever the user asks for the drivers) or keep the sales end and mark sessions / conversion / buy-box for the days after the traffic end as provisional in the output. State which you did.
- Gap days = any
dateinside a window below the threshold, including mid-window holes, not only the lagging tail - flag and exclude them. End both windows at the window end found above (the traffic window end when you report drivers).
- Pull both windows per day and per ASIN from
amazon_profit_by_sku_and_date(CSV,limit: 5000,from/tomandatory), filtered to the scanned marketplace (marketplace_country_code, as in the step-3 probe) so the per-day ASIN counts in step 5 are for one marketplace:groupBy [date, child_asin, product_name](addmarketplace_country_codeorcurrencytocolumns+groupByif you need them in the rows - they are not emitted otherwise) with aggregationssum(total_sales),sum(total_units_sold),sum(total_orders), andmax(total_sessions),max(total_page_views),max(avg_buybox_percentage)-max, notsum, because those three repeat on every SKU row of the ASIN (see Configuration);maxcollapses the duplicates to the ASIN's single daily value. Grouping on[date, child_asin]rather than the ASIN alone is what keeps that dedupe exact before you add across days. Row budget: 7 days x ~1,600 ASINs is ~11k rows per window, over the 5,000-row CSV cap, so one export per window is not enough. Split each window by date into 2-day chunks (4 exports per window - 2+2+2+1 days; measured ~4,300 rows each on a ~2,150-row/day account, 86% of the cap; 8 exports in total, fired concurrently and polled together). If a chunk returns more than ~4,500 rows orrowCountequal to the limit, switch that window to single-day chunks (~2,150 rows each) or page withskip: 5000on the same query (sameorderByColumn), and never treat a chunk that hit the cap as complete. Clean the rows first. Some SKU rows of an ASIN come back with a blankproduct_name, so the[date, child_asin, product_name]grouping can split one ASIN-day into two rows (observed: 397 ASIN-days split; the blank-name rows were all zeros), and a few rows have a blankchild_asin. Drop the rows with a blankchild_asin, then re-collapse by[date, child_asin]in code: sum sales, units and orders; take the max of sessions, page views and buy box; keep the non-blank name. Then aggregate per ASIN per window in code: sumtotal_sales,total_units_sold,total_sessionsandtotal_page_viewsacross the window's days (total_orderscounts a multi-ASIN order once per ASIN, so never present its sum as the account order count); buy box =avg_buybox_percentageaveraged withtotal_page_viewsas the weight, leaving out ASIN-days with 0 page views - an ASIN-day with no traffic data carries a buy box of 0, so a simple average drags every SKU down. Use the simple average only as a fallback when the ASIN has no page views at all in a window, and say so. Conversion = units / sessions; price = sales / units. Also count, perdate, the distinctchild_asinwith a row (the sales ASIN count) and those withtotal_sessions > 0(the traffic ASIN count) - step 5 needs both, and the traffic count re-checks the step-3 probe. - Check both windows for completeness (see the guard): compute each window's
effective days twice - for sales, sum over its days of (day's sales ASIN count /
median daily sales ASIN count across both windows); for traffic, the same with the
traffic ASIN count (ASINs with
total_sessions > 0). If the two windows' effective days differ by more than 0.3 days on either count, or either has gap days, the comparison is apples-to-oranges - do not report the delta as real (observed: rows gave 6.75 vs 7.03 effective days, which looked fine, while ASINs with traffic gave 6.01 vs 7.01 - the recent window's last day had no per-ASIN traffic at all). Shift rule: move both windows back, keeping them adjacent and equal, so the recent window ends on the last day of the latest contiguous complete run from step 3 - sales-complete, and traffic-complete per ASIN when you report drivers (e.g. 09-19..09-25 vs 09-12..09-18 at 2.9 vs 6.9 effective days -> 09-14..09-20 vs 09-07..09-13 at 6.9 vs 6.9), re-pull step 4 and recompute. Report the shift in the output header ("windows shifted back to end {date}: later days incomplete"). Only if no complete pair exists, flag the scan provisional and say which window is under-reported. Unshipped carry-over. Sum the step-3 per-dayunshipped_salesover each window. The per-day 10% rule can pass both windows and still leave the recent one with more orders waiting to ship (observed: 694.84 more unshipped, 0.94% of the recent window's sales, enough to move the net change from -3,287 to about -2,592). When the recent window's unshipped minus the prior window's is more than 0.5% of the recent window's sales, add the carry-over line to the output header (Y = net change + that difference). - Join on child ASIN and compute per ASIN: sales change (abs + %), and the change in sessions, conversion (units / sessions), price-per-unit and buy-box.
- Rank by absolute sales change; take the top gainers and top decliners.
- Decompose each to its dominant driver (traffic / conversion / price / buy-box, or n/a / availability) with the log split from the framework and attach the lever. Add the catalog roll-up (concentrated vs broad).
Output format
Sales Movers - {marketplace} - {recent window} vs {prior window} (ends {last complete day}; both windows complete)
{if shifted: "Windows shifted back to end {date} - data after it is incomplete (lag {n} days)."}
{if the recent window includes days after the traffic window end: "Sessions / conversion / buy-box for {dates} are provisional - Amazon traffic data lags sales by {n} days here and may be revised."}
Net catalog change: {cur}{delta} ({pct}%) · concentrated in {k} products / broad
{if recent-window unshipped exceeds the prior window's by > 0.5% of recent sales: "{cur}{X} more is still unshipped in the recent window; if it ships, the net change is about {cur}{Y}."}
TOP DECLINERS (by money lost)
Product / ASIN sales (was -> now) driver detail lever
{product name} / {ASIN} {cur}.. -> {cur}.. traffic -{x}% sessions {a}->{b} rank/ads/suppression
{product name} / {ASIN} {cur}.. -> {cur}.. buy-box -{x}pp bb {a}%->{b}% pricing/competitor
{product name} / {ASIN} {cur}.. -> {cur}.. conversion -{x}% CVR {a}%->{b}% listing/price/reviews
TOP GAINERS (by money gained)
{product name} / {ASIN} {cur}.. -> {cur}.. traffic +{x}% sessions {a}->{b} protect / scale ads
{product name} / {ASIN} {cur}.. -> {cur}.. conversion +{x}% CVR {a}%->{b}% listing/offer win - protect
{product name} / {ASIN} {cur}.. -> {cur}.. buy-box +{x}pp bb {a}%->{b}% offer won back - hold price/stock
Biggest single swing: {product name} / {ASIN} {cur}{delta} - {driver} -> {lever}Buy box (bb) is the page-view-weighted average; the driver is the largest log-split term in the direction of the move.
Worked example (illustrative)
A weekly scan shows the catalog down modestly, but the decline is concentrated in two
SKUs, not broad. The first: sales down a third while sessions held - conversion fell,
and its avg_buybox_percentage dropped from ~95% to ~40% -> buy-box loss is the cause
(a reseller/price move), routed to pricing, not copy. The second: sales down with
sessions down the same amount -> a traffic problem (check rank/ads/suppression), not
the listing. Meanwhile a gainer doubled on rising sessions -> ads/rank working, protect
it. The output is a short, ranked, already-diagnosed list - not a spreadsheet of every
SKU's delta.
Quality self-check
- Did I rank by absolute money change, not percent (so hero SKUs surface)?
- Did I decompose each mover into traffic vs conversion vs price vs buy-box with the log split, so every mover has a defined driver (or n/a / availability), not just the delta?
- Did I check buy-box on every decliner (the classic hidden conversion killer), weighting it by page views and leaving out ASIN-days with 0 page views?
- Did I label each row by product name / ASIN (the unit the skill measures), after re-collapsing blank-name and blank-ASIN rows?
- Did I detect the last complete day dynamically (not assume a fixed lag), anchor at the end of a contiguous complete run, and check BOTH windows for completeness / equal effective days before trusting the delta - shifting both back and reporting it if not?
- Did I take
max(), notsum(), for sessions / page views / buy box when grouping SKU rows to the ASIN, and compute conversion as units / sessions (there is no rate column)? - Did I judge traffic completeness separately from sales and per ASIN, not only at account level (measuring the lag, not assuming it), end the windows at the per-ASIN traffic end when reporting drivers, and compare traffic effective days within 0.3 days?
- Did I add the unshipped carry-over line when the recent window has more than 0.5% of its sales still waiting to ship compared with the prior window?
- Did I say whether the move is concentrated or broad (one SKU vs seasonality)?
- Did I keep each marketplace in its own currency?
Common mistakes
- Reporting deltas with no cause - the decomposition (traffic/conversion/price/buy-box) is the whole point.
- Ranking by percent - tiny SKUs dominate and the real money hides.
- Comparing windows of unequal completeness - a partial recent week fakes a drop, a partial prior week fakes a boom. Check effective days on both sides.
- Blaming the listing when sessions fell (that's traffic) or when the buy-box dropped (that's pricing/competitor).
- Summing sales across marketplaces/currencies into one number.
- Reacting to a single SKU's noise instead of the ranked, material movers.
- Reporting a broad, uniform, all-traffic move (either direction) as real - that is an incomplete window on one side; shift both windows back to the end of the latest contiguous complete run and re-pull, or flag it - don't cry wolf.
- Anchoring the windows to an isolated complete day that sits after a hole (a 3/7 recent week reads as -58%), or pulling a whole 7-day window per ASIN per day in one export and counting ASINs from a file truncated at the 5,000-row cap - chunk the window by date.
- Summing
total_sessions/total_page_views/avg_buybox_percentageacross the SKU rows of one ASIN - they repeat per SKU, so multi-SKU ASINs double-count; group by[date, child_asin]and takemax(). - Reading a sessions drop on the days after the traffic-complete tail as a traffic move - that is the Amazon traffic-report lag; sales complete ~1 day behind, traffic by a measured 2-3 to 8-13 days, and per-ASIN traffic a day or more after the account total.
- Trusting the account-level
total_sessionsalone for the traffic end - it can be full on a day when no child ASIN has sessions yet, and every product then reads as a traffic drop. Count the ASINs with sessions per day. - Averaging buy box without page-view weights - ASIN-days with no traffic data carry a buy box of 0 and pull every SKU down.
- Calling a buy-box/conversion rise from ~0 an organic win when it's back-in-stock, or reporting a >100% conversion rate literally (units-per-session quirk).
Notes
- Read-only (analysis). Fixes that follow (price, listing, ads) are separate write skills, each dryRun-gated.
- Complements the weekly business review (account-level) and the buy-box root-cause skill (buy-box only) - this is the catalog-wide, cause-attributed mover scan.
- A DataDoe skill, built on DataDoe's Profit by SKU & Date data (sales, sessions, page views, buy-box per child ASIN per day) with Profit by Date for completeness calibration.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

