Skip to content

feat(browser): add "don't build faster than you can test" rule to byg-et-flow - #6

Merged
simo787c merged 1 commit into
mainfrom
feat/agentbase-browser-tracer-bullet-rule
Jul 27, 2026
Merged

feat(browser): add "don't build faster than you can test" rule to byg-et-flow#6
simo787c merged 1 commit into
mainfrom
feat/agentbase-browser-tracer-bullet-rule

Conversation

@simo787c

Copy link
Copy Markdown
Contributor

Follow-up to #5. That PR merged at the moment its branch held only the first commit (the stepped interview), so the tracer-bullet build-discipline rule got left behindmain shipped the new interview but not this rule. This PR lands the missing commit.

The rule

A real bilagskontrol run interviewed well but then built and saved the full match-flow on a single faktura (no betalingsdokumentation, no Excel-oversigt), so the matching could never run and the flow was left "gemt, men uprøvet". It also guessed an undocumented node config instead of probing it.

Adds to byg-et-flow's build loop: you need a sample you can run the flow helt igennem on before building the whole graph; with a partial sample, build and verify only the runnable slice now (e.g. OCR + sortering on the one bilag via run_node) and defer the rest; probe uncertain node config on real data with run_node instead of guessing; never save a full flow as "færdigt" when it has never run.

RED/GREEN

On the exact "one faktura, rest comes later, fortsæt" decision point: a faithful (non-leading) RED with the interview-only skill reproduced the failure — it laid the whole 6-node graph in one set_flow and saved it v0.1 untested, following "læg hele grafen i ét set_flow" literally, only then noticing it couldn't run. GREEN built only the OCR slice, ran it on the real faktura via run_node (surfacing that the filename key I2 differs from fakturanr 100447), and deferred the matching until a complete sag exists — saving nothing untested.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SYvwyQEa9xsKzsQTwpBuhp

…-et-flow

A real bilagskontrol run of the stepped-interview skill interviewed
well but then built and SAVED the full match-flow on a single faktura
— no betalingsdokumentation, no Excel-oversigt — so the matching logic
could never be run and the flow was left "gemt, men uprøvet". It also
guessed an undocumented node config instead of probing it.

Add a tracer-bullet rule to the build loop: you need a sample you can
run the flow HELT IGENNEM on before building the whole graph; with a
partial sample, build and verify only the runnable slice now (e.g. OCR
+ sortering on the one bilag via run_node) and defer the rest; probe
uncertain node config on real data with run_node instead of guessing;
never save a full flow as "done" when it has never run.

RED/GREEN tested with fresh subagents on the exact "one faktura, rest
comes later, fortsæt" decision point. Faithful (non-leading) RED with
the interview-only skill reproduced the wild failure: it laid the whole
6-node graph in one set_flow and saved it as v0.1 untested, following
"laeg hele grafen i et set_flow" literally, only then noticing it could
not run. GREEN with the rule built only the OCR slice, ran it on the
real faktura via run_node (surfacing that the filename key I2 differs
from fakturanr 100447), and explicitly deferred the matching until a
complete sag exists — saving nothing untested.

Co-Authored-By: Claude <noreply@anthropic.com> via Dash <dash@syv.ai>
Claude-Session: https://claude.ai/code/session_01SYvwyQEa9xsKzsQTwpBuhp
@simo787c
simo787c merged commit 5b88642 into main Jul 27, 2026
1 check passed
@simo787c
simo787c deleted the feat/agentbase-browser-tracer-bullet-rule branch July 27, 2026 12:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant