Refine: one name for the API key, one window for usage, and the fixes a real box turned up - #1
Merged
Merged
Conversation
v0.1.0 accepted the key as either BWG_API_KEY or BWG_KIWIVM_API_KEY.
The code called the short one canonical; every README, skill file,
installer line and error message taught the long one. So the pair a
new user copied read
export BWG_VEID=1347645
export BWG_KIWIVM_API_KEY=private_xxx
— two halves of one credential in two styles, one of them namespaced
twice, with nothing in the tool saying which was real.
BWG_VEID and BWG_API_KEY now read as the pair they are. The old
spelling still authenticates, is documented in one README line, and
warns once per run on stderr naming the replacement; it goes at v1.0.
`bwg server show env` reports whichever variable is actually in play,
because a tool that will not say where its credentials came from is
how this got confusing in the first place.
BREAKING CHANGE: none yet — BWG_KIWIVM_API_KEY keeps working until
v1.0. Rename it now and the warning goes away.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three things `bwg info` and its neighbours got wrong on a real box: - KiwiVM returns one PTR entry per address whether or not a record is set, and Tabbed drops empty values, so `bwg info` printed a bare "rDNS" heading with nothing under it — which reads as data that failed to load. output.Section prints a heading only when something survives beneath it; Bandwidth, rDNS and Health all use it. - A box suspended at 100% bandwidth with zero abuse points was told "see: bwg abuse", which correctly answered "nothing outstanding". A dead end at the moment the answer mattered. It now reads "yes — transfer quota exhausted (the usual cause); resets in 30d", phrased as the inference it is, and still points at bwg abuse when there are abuse points or a violation on the record. - "1 keys", "1 points", "%d server(s)". output.Count(n, noun) means no message has to hedge with (s). Also: colour detection is now a pure function over (env, isTTY) so the NO_COLOR / BWG_COLOR / TERM=dumb precedence is testable without a pty, and `bwg run` checks whether the stream it is about to read is a terminal rather than asking os.Stdin on principle. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
`bwg usage` defaulted to every sample KiwiVM kept — 600-odd rows,
oldest first, which put the recent end of the series and the quota
summary off the bottom of the screen.
The defect underneath was worse: --days 7 trimmed the table but not
the arithmetic, so seven rows sat above "Total: 5.9 TiB ... over
608d", and --raw ignored the window entirely. Three parts of one
screen describing three different spans, each individually correct.
One trim now feeds the table, --raw, the totals line and the JSON
payload, so they cannot disagree. The default window is 30 days —
the billing cycle the quota line underneath it is measured against —
and the render says what it withheld:
Showing 30 of the 608 days KiwiVM kept — for all of it: bwg usage --days 0
JSON gains "window":{"days","available"}; "totals" now covers the
window rather than the whole series.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Download writes the new binary to $TMPDIR, which is a separate filesystem from $HOME on most Linux boxes. os.Rename cannot cross that boundary, so `bwg update` reported a failed install on a perfectly healthy update — for anyone whose /tmp is a tmpfs. Replace now falls back to copying into the target directory when the rename fails, and still restores the old binary if the install does not complete. Tests cover Download's asset selection, both archive formats, the copy fallback and the restore path; the zip branch had never been executed at all. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
internal/cli tests every command for breadth; the MCP run closures are a second execution path and nothing called them. Coverage there went 56.6% -> 86.4%, which is mostly "these fifteen tools do not panic and return JSON an agent can read". Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
TASTE.md gains the scars for the credential-name decision and the usage window. SKILL.md and CLAUDE.md follow the JSON shape and the new output helpers, so agents are not reading a second contract. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Six commits from running
bwgagainst a live box that had just comeback from a bandwidth suspension. Every change here is something that
box exposed.
What changed
BWG_API_KEYis the name. v0.1.0 accepted two spellings and thecode and the docs each picked a different one, so the pair people
copied was
BWG_VEID+BWG_KIWIVM_API_KEY. The old name keepsworking, warns once per run on stderr, and goes at v1.0.
bwg infoprinted a barerDNSheading with nothing under it —KiwiVM returns an entry per address whether or not a PTR is set.
output.Sectionprints a heading only when something survives underit.
A box suspended for bandwidth was told
see: bwg abuse, whichanswers "nothing outstanding". It now names the likely cause and when
the quota resets, and still routes to
bwg abusewhen there isactually an abuse case.
bwg usagedumped 608 rows and--days 7trimmed the table butnot the totals, while
--rawignored the window entirely. One trimnow feeds table,
--raw, totals and JSON; default 30 days; the rendersays what it withheld.
bwg updatewas broken wherever$TMPDIRis a different filesystemfrom the binary (tmpfs
/tmp): the install rename failsEXDEV.Falls back to a copy.
Plus
1 keys→1 keyand friends.Verification
Coverage 79.0% → 84.5%. mcp 56.6→86.4 (every tool actually runs now),
updater 50.0→79.2 (Download, the zip branch and the EXDEV fallback had
no tests at all), output 75.5→83.8, cli 77.8→81.8.
Design rulings behind the two judgement calls are in
TASTE.md.🤖 Generated with Claude Code