sparebank1-context

Name: sparebank1-context
Author: magnusrodseth

22stars0forks

How to install sparebank1-context?

magnusrodseth/sparebank1-cli/sparebank1-context

$npx -y skills add magnusrodseth/sparebank1-cli --skill sparebank1-context

Installs into the current project.

Summary

Build and maintain a personal finance context for the sb1 CLI (SpareBank 1) by running a short onboarding interview over the user's own accounts and transactions, then saving the answers to a private, git-ignored file the agent reads at the start of future finance sessions. Cap

Files · 1

View on GitHub

SKILL.md

1	# sparebank1-context
2
3	Turn `sb1` from "shows transactions" into "reasons over your money". This skill
4	runs a brief interview over the user's real data, then stores the answers as a
5	personal finance context that future finance sessions load up front. Read
6	[sparebank1-shared](../sparebank1-shared/SKILL.md) first for auth and flags.
7
8	The CLI already surfaces the data. This skill captures the *personal meaning
9	layer* the bank cannot infer, and nothing more.
10
11	## The privacy contract (read this first)
12
13	This skill's mechanism is public (it ships in a public repo and installs for
14	anyone), but the answers are private personal data. Therefore:
15
16	- Write the context file to a user-owned path outside this repo. Default:
17	`~/.config/sparebank1-cli/context.md` (alongside where the `file` secret store
18	writes). Override with the `SB1_CONTEXT_FILE` environment variable.
19	- Never write answers, balances, account numbers, or counterparties into
20	this repository, a commit, a PR, or any shared/cloud-synced location. Treat the
21	file like the plaintext `file` secret store: keep it off backups and out of
22	anything that syncs.
23	- Prefer account names + last 4 digits over full account numbers. Don't
24	record figures the user didn't ask you to keep.
25	- The file is for this user's agent sessions. Don't paste its contents into
26	chats, issues, or external services.
27
28	Resolve the path once at the start:
29
30	```bash
31	ctx="${SB1_CONTEXT_FILE:-$HOME/.config/sparebank1-cli/context.md}"
32	mkdir -p "$(dirname "$ctx")"
33	```
34
35	## When to run
36
37	- First time: no context file exists → run the full interview below.
38	- Refresh: the file exists → read it, show the user what's captured, and only
39	ask about what's new or changed (new counterparties, new Uncategorized rows,
40	subscriptions to revisit). Don't re-ask settled questions.
41
42	Confirm login before pulling data:
43
44	```bash
45	sb1 status # if not logged in, ask the user to run `sb1 login` (see sparebank1-shared)
46	```
47
48	## Step 1 — gather the data (no code change needed)
49
50	Drive the interview from existing commands. Start with a short window:
51
52	```bash
53	sb1 --json summary --months 3
54	```
55
56	From the JSON, the rows worth asking about are:
57
58	- `categories[]` — especially the `"Uncategorized"` bucket (the bank
59	couldn't classify these; you can, with the user's help).
60	- `topOutgoing[]` — the biggest counterparties by spend.
61	- `subscriptions[]` — what the bank already flagged as recurring.
62
63	To see the actual Uncategorized rows behind that bucket, drill in per account:
64
65	```bash
66	sb1 --json transactions -a <account> --from <start> --to <end> --classified
67	```
68
69	Each row has `category`, `recurring`, `subscription`, `counterpartyName`,
70	`amount` (negative = outgoing), and `date`. Filter to rows with a missing/`"Uncategorized"`
71	category or to the top counterparties — those are your interview targets.
72
73	## Step 2 — the interview (keep it short)
74
75	Prioritize the `Uncategorized` bucket and the top counterparties. Do not walk
76	every transaction. Group similar rows and ask once per group. Aim for a handful
77	of focused questions, not a questionnaire. Capture these dimensions:
78
79	\| Dimension \| The question to resolve \|
80	\|---\|---\|
81	\| Income classification \| For each inbound counterparty: salary, reimbursement, transfer from a partner, or a refund? (So "income" isn't overcounted.) \|
82	\| Work-expensed / reimbursed \| Which outgoing rows does work pay back (travel, equipment)? These should net OUT of "real" spending. \|
83	\| Account semantics \| What is each account for — daily, buffer, a savings goal, BSU, joint? \|
84	\| Shared / split \| Anything split with a partner where only the user's share should count? \|
85	\| Counterparty aliases \| Cryptic strings (e.g. `KLARNA*…`, acquirer codes) → a human label. \|
86	\| Subscriptions \| For each flagged (or user-known) subscription: keep, cancel, or annual? \|
87	\| Fixed vs discretionary \| Which spend is fixed (rent, loan, insurance) vs discretionary? \|
88
89	Ask in plain language, tied to what you actually saw: "You sent ~kr 4 000/month
90	to `BERG EIENDOM` — is that rent (fixed)?" beats an abstract checklist. Skip
91	dimensions that don't ap

People who liked this also viewed

6 picks

Type

sparebank1-context

22stars0forks

View on GitHub

How to install sparebank1-context?

magnusrodseth/sparebank1-cli/sparebank1-context

$npx -y skills add magnusrodseth/sparebank1-cli --skill sparebank1-context

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# sparebank1-context
2
3	Turn `sb1` from "shows transactions" into "reasons over your money". This skill
4	runs a brief interview over the user's real data, then stores the answers as a
5	personal finance context that future finance sessions load up front. Read
6	[sparebank1-shared](../sparebank1-shared/SKILL.md) first for auth and flags.
7
8	The CLI already surfaces the data. This skill captures the *personal meaning
9	layer* the bank cannot infer, and nothing more.
10
11	## The privacy contract (read this first)
12
13	This skill's mechanism is public (it ships in a public repo and installs for
14	anyone), but the answers are private personal data. Therefore:
15
16	- Write the context file to a user-owned path outside this repo. Default:
17	`~/.config/sparebank1-cli/context.md` (alongside where the `file` secret store
18	writes). Override with the `SB1_CONTEXT_FILE` environment variable.
19	- Never write answers, balances, account numbers, or counterparties into
20	this repository, a commit, a PR, or any shared/cloud-synced location. Treat the
21	file like the plaintext `file` secret store: keep it off backups and out of
22	anything that syncs.
23	- Prefer account names + last 4 digits over full account numbers. Don't
24	record figures the user didn't ask you to keep.
25	- The file is for this user's agent sessions. Don't paste its contents into
26	chats, issues, or external services.
27
28	Resolve the path once at the start:
29
30	```bash
31	ctx="${SB1_CONTEXT_FILE:-$HOME/.config/sparebank1-cli/context.md}"
32	mkdir -p "$(dirname "$ctx")"
33	```
34
35	## When to run
36
37	- First time: no context file exists → run the full interview below.
38	- Refresh: the file exists → read it, show the user what's captured, and only
39	ask about what's new or changed (new counterparties, new Uncategorized rows,
40	subscriptions to revisit). Don't re-ask settled questions.
41
42	Confirm login before pulling data:
43
44	```bash
45	sb1 status # if not logged in, ask the user to run `sb1 login` (see sparebank1-shared)
46	```
47
48	## Step 1 — gather the data (no code change needed)
49
50	Drive the interview from existing commands. Start with a short window:
51
52	```bash
53	sb1 --json summary --months 3
54	```
55
56	From the JSON, the rows worth asking about are:
57
58	- `categories[]` — especially the `"Uncategorized"` bucket (the bank
59	couldn't classify these; you can, with the user's help).
60	- `topOutgoing[]` — the biggest counterparties by spend.
61	- `subscriptions[]` — what the bank already flagged as recurring.
62
63	To see the actual Uncategorized rows behind that bucket, drill in per account:
64
65	```bash
66	sb1 --json transactions -a <account> --from <start> --to <end> --classified
67	```
68
69	Each row has `category`, `recurring`, `subscription`, `counterpartyName`,
70	`amount` (negative = outgoing), and `date`. Filter to rows with a missing/`"Uncategorized"`
71	category or to the top counterparties — those are your interview targets.
72
73	## Step 2 — the interview (keep it short)
74
75	Prioritize the `Uncategorized` bucket and the top counterparties. Do not walk
76	every transaction. Group similar rows and ask once per group. Aim for a handful
77	of focused questions, not a questionnaire. Capture these dimensions:
78
79	\| Dimension \| The question to resolve \|
80	\|---\|---\|
81	\| Income classification \| For each inbound counterparty: salary, reimbursement, transfer from a partner, or a refund? (So "income" isn't overcounted.) \|
82	\| Work-expensed / reimbursed \| Which outgoing rows does work pay back (travel, equipment)? These should net OUT of "real" spending. \|
83	\| Account semantics \| What is each account for — daily, buffer, a savings goal, BSU, joint? \|
84	\| Shared / split \| Anything split with a partner where only the user's share should count? \|
85	\| Counterparty aliases \| Cryptic strings (e.g. `KLARNA*…`, acquirer codes) → a human label. \|
86	\| Subscriptions \| For each flagged (or user-known) subscription: keep, cancel, or annual? \|
87	\| Fixed vs discretionary \| Which spend is fixed (rent, loan, insurance) vs discretionary? \|
88
89	Ask in plain language, tied to what you actually saw: "You sent ~kr 4 000/month
90	to `BERG EIENDOM` — is that rent (fixed)?" beats an abstract checklist. Skip
91	dimensions that don't ap

People who liked this also viewed

6 picks

Type