Repository navigation
Release v0.3.0 - #367
Merged
Merged
Release v0.3.0#367
Conversation
…tures show one The README opened on Kitchens and the agent, and its pictures were screens every recipe app has. It now opens on what a comparison with nine other self-hosted recipe apps found only here: every version of a recipe kept, each cooking recorded with what was changed at the stove, and another cook's version read beside your own. - Three new pictures, taken from a throwaway 0.2.0 instance holding a staged ratatouille: its History on a phone, with a translation; a cooking with the lines that changed; and Mum's version read from yours on a computer. - "What you get" starts with the history, the diary and Kitchens. - "Who it is for" names the cook who changes a recipe each time, and says there is no weekly plan, no pantry and no shared shopping list. - The Docker Hub page and its one-line description say the same. They reach Docker Hub with the next release. docs/readme/wide-recipe.png is no longer shown. It stays until a release has published the new Docker Hub page, since the page now live reads it from main. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…ng in two of them The commit before this one put eight pictures at the top, dropped the recipe with its photograph from first place, and rewrote more of the README than was asked. This brings it back to the layout it had: - The pictures are the six there were, in their places. On the phone, History stands where the recipe was and a cooking with its changed lines where Shopping was. Both are the height of the others. - The History holds a Translation. - The opening says, in order, that every version is kept, that everyone writes in their own Cookbook, and that an agent can do what a person can. The rest of the README is as it was. docs/readme/recipe.png and shopping.png are no longer shown. They stay until a release has published the new Docker Hub page, since the page now live reads them from main. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
docs: the README opens on every version being kept, and two of its pictures show it
A Translation's chain starts fresh, so the History found its split before the first Version and drew a band there saying the recipe split. Each row of a Translation now sits under the Version it translates, on a curve out of that Version's dot, whenever it was saved. A Copy or a variation leaves the last Version it shares the same way, and the band at a split is gone. A Branch with no Version of its own gets a hollow-dot row under the Version it starts from. The recipe a variation outgrew gets one under its last Version, saying nothing changed since. Both were left out before. Two line colours join the three, so five Branches are told apart. Closes #250 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…line The History draws every line out of the dot it started from (Closes #250)
A Version cooked three times and rated "Again" each time showed "Again" three times. Now the reader's own cookings are one badge, the pot and the last rating they gave that Version, and everyone else's are a second, dashed badge naming who cooked. A badge for one cooking opens it. A badge for several opens the list of them, newest first. get_thread says whose each Attempt is in two new fields, person_name and mine. Closes #249 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…badges feat: the History shows a Version's cookings as two badges at most
…ing it damaged A Translation's chain starts fresh (ADR 0006), so it shares no Version with the recipe it translates, nor with another Translation of it. Both Operations walked the two chains back, found nothing in common and answered an internal error calling a healthy recipe a damaged Bundle. They now share one helper for the last Version in common. When there is none and either Branch is a Translation, the call is refused as a bad request that names where the link is held: translates_version_id on the Translation's Versions, and translation on its Branch in get_thread and get_recipe. Two Branches that share no Version and translate nothing are still reported as damaged. Both summaries in the Catalogue say a Translation has no Branch Point with its source, so an agent reads it before asking. Closes #251 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…oint fix: branch_point and divergence refuse a Translation instead of calling it damaged (Closes #251)
The picture was taken before #250. It showed a band reading "Splits into 2 Branches here" above the first Version and the French Translation at the foot of the page. Retaken on an image built from dev, on the same staged recipe and in the same 390x844 frame: each line leaves the dot of the Version it started from, and the Translation sits under the Version it translates. Closes #253 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
docs: the README's History picture shows the History as it is drawn now
Records the decision on AI-assisted recipe import. Kamosu ships a Runner that asks it for Jobs and has an agent do each one, under a narrow Runner Key. Kamosu itself calls no model and holds no provider's key. - ADR 0048, with the in-app model call as the option set aside - Glossary: Runner, Runner Key and Original, and the entries they touch - Research: how other apps read recipes with a model, and a test of the in-app call - A dated correction to the older note that said no recipe app reads pages without structured data Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
docs: an agent beside Kamosu reads what Kamosu cannot (ADR 0048)
Three reviews of the spec (#258) found the first version wrong in places. - The agent no longer holds the Runner Key. Each Job comes with a Job Key that acts as the asker, only for that Job, and is spent when it ends. - A Job's write never makes a Copy, so an asker who left the Cookbook fails the Job where the first version claimed a refusal that did not exist. - Deleting a stray Food is gone from the list: only an Operator deletes a Food and nothing records which save created one. - A link whose fetch is refused now lands bare and goes to the agent. - An Original belongs to the Job that reads it, not to the Import. - ADR 0033 names the Runner as the one exception to the single client. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
docs: rewrite ADR 0048 around a key for each Job
…ing else A Runner Key is the first caller Kamosu knows that is not a Person (ADR 0048). An Operator signed in directly mints one, lists them by name and last use, and revokes one on its own. The Secret is shown once and stored hashed. The Catalogue gains a fourth level of permission, Runner, and one Operation declared with it: ask_for_work, which answers that nothing is waiting until #262 hands over a Job. A Runner states its version and a Kamosu of another version refuses it, naming both. Whoever asks is now one of three things, nobody, a Person or a Runner, held as one type. An Operation written for a Person takes a Person, so a Runner Key cannot reach one by mistake. The Core's one check refuses a Runner Key every other Operation, the ones open to anyone included, and every route outside the Catalogue. The MCP door lists it ask_for_work alone, and lists that to nobody else. runner_status tells any Person whether the instance has a Runner and when one last asked for work. Closes #260 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
An Operator mints a Runner Key, which can ask for work and nothing else (Closes #260)
While a Runner Key stands, a page with no structured recipe data, or one that refused the fetch, lands as a bare recipe and leaves a Job for an agent on it (ADR 0048). No Runner takes the Job yet: that is #262. The Job is a row in `jobs` under the operation `read_by_agent`, marked as an agent's by a row in the new `agent_jobs` table (migration 43). That mark keeps it out of both lanes and out of the recovery at startup, so a restart leaves it waiting. A recipe has one such Job at a time, and deleting the recipe ends it with the reason. The Import Report names the Job in `agent_jobs`, and `list_imports` shows it inside its arrival with its state now. `fetch_page` tells a page's own refusal from a link nobody answered; 404 and 410 stay refused. Closes #261 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
A link Kamosu cannot read lands bare and a Job waits for an agent (Closes #261)
A Runner asking for work is handed the oldest Job that waits for an agent, with the link to read and the recipe it is for (ADR 0048). The Job is running from then on, held by that Runner Key and no other. A Runner holds one Job at a time, so asking again answers the Job it holds. A restart leaves a held Job held. report_on_job is the Runner's second Operation. It ends the Job as completed, failed, or could not be read, and passes on the agent's own report: what it left undone, what it was unsure of, likely duplicates and the Foods its corrections left unused. Only the Runner Key holding a Job reports on it, and any other is answered as if the Job were not there. An Import Report is kept as it was written, and each reading adds the agent's Job as it stands now, so the bare recipe and the agent's work stay one arrival. A completed Job says an agent made the recipe. A failed one is listed under what could not be read, with the reason, and the bare recipe stays listed as arrived. Migration 44 adds to agent_jobs which Runner Key holds the Job, how it ended and the agent's report. Closes #262 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…-reports A Runner takes a waiting Job and reports how it ended
Taking a Job hands the Runner a Job Key with it, for the agent that does the Job (ADR 0048). It is a Secret: Kamosu keeps its hash, shows it once, and it names the asker only while the Job is running. A Runner that asks again is handed a new one, and the one before stops working. What a Job Key may call is a new mark on every Operation's declaration, `job_key`. Nine reads carry `JobKey::Reads`: get_recipe, get_thread, branch_point, divergence, changed_since, search_recipes, list_tags, list_foods and get_food. The Core's one check meets a Job Key first and answers everything else as an Operation that does not exist, the ones open to anyone included. After that the call passes the ordinary checks as the asker. Every route outside the Catalogue answers a Job Key 404, and the MCP door lists it its nine tools alone. One test holds the list by name, so marking another Operation fails the build until the name is added on purpose. Closes #263 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
The Job Key reads the library as the Person who asked
A Job Key now writes as well as reads. Five Operations join its list: save_recipe_version, edit_recipe, set_reading, set_recipe_tag and upload_photograph, with POST /api/photographs as the one route outside the Catalogue it may use. The Core holds each write before the Operation runs, so no Operation learns about Jobs. A write aimed at any recipe but the Job's answers as a recipe that is not there. Where the asker's own change would start a Copy, the write is refused and the Job fails with the reason, which spends the Job Key. A save names no translates_version_id, and a Reading is not corrected while another recipe holds the same Version. Closes #264 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
The Job Key writes the one recipe of its Job
A save writes down which agent's Job wrote a Version, beside the Access Key (migration 46). It is local: in no fingerprint and no Bundle. A Job's saves fold into the one Version that Job wrote, however far apart, and never into a Version from before the Job. A save by a Session or an Access Key never folds into a Job's Version, on the recipe or on a Copy of it, and folds as before everywhere else. Once a Job has saved, its next save is refused and the Job fails when somebody else saved the recipe since, or when a Copy, a variation or a Translation holds the Version it was shaping. The MCP door's instructions said a collapsed save erases the name and change note it does not send, which #165 made untrue. They say what the code does now. Closes #265 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…ersion A Job leaves at most one new Version (Closes #265)
A held Job has a deadline two minutes ahead, and the Runner moves it on with the new Operation `renew_hold`. A Job whose deadline passed waits again and goes to the next Runner that asks. The Runner that lost it is refused when it renews or reports, and told the hold is gone. The deadline is stored, so it survives a restart. A Job is tried three times. A hold that lapsed and a report of `failed` each end a try, and the third ends the Job as failed with a reason that names the limit. `could_not_be_read` ends it at once. `ask_for_work` takes a mark, `ask`. The same ask sent again replaces the Job Key of the reply that was lost, and a new ask spends none. Job Keys move to a table of their own, so the one from a lapsed hold works beside the next Runner's until the Job ends. Revoking a Runner Key sends its Job back to waiting and ends the Job Keys it was handed. A disabled asker's waiting Job is handed to nobody, and a deleted asker's Jobs end as failed. Closes #266 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…s-hold A Runner keeps its hold by renewing it (Closes #266)
cancel_job ends an agent's Job whether it waits or a Runner holds it. One of Kamosu's own is still cancelled only while it waits in line. Cancelled is an end like the others, so the Job's Job Keys answer as unknown Secrets from the cancel on, and what the agent saved stays. The Runner holding the Job is refused its next renew_hold and told the Job was cancelled, and a report that arrives afterwards is refused in the same words. Kamosu never calls the Runner. A cancel counts as no try. Only the Person who asked cancels. That is asked by jobs::ensure_asker, apart from ensure_reader, so that letting a Co-author read an agent's Job (#268) lets nobody new cancel one. Closes #267 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…-place fix: a line moved without rewording keeps each page's own order (closes #337)
Every release publishes agent-image/ as battermanz/kamosu-agent, under Kamosu's own version tags. The agent's Dockerfile takes VERSION and GIT_SHA as Kamosu's does, through the same stamp, so its revision label says which agent-image/ an image holds. The release hook reads that to build the agent only when agent-image/ changed, and gives an unchanged release's tags to the image already published. Alpine's security fixes reach an unchanged image through the rebuild line, a date inside the packages' RUN step, which a new checklist item asks the release to move. It is in the RUN and not a comment, since a build reuses a step whose text has not changed. The agent has its own Docker Hub page, held to Docker Hub's limits and to the Runner guide's Compose file. The Runner guide, the README, the notices and ADR 0048 now say Kamosu publishes the agent, and how to mount your own instructions over the image's. Closes #345 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
Every release publishes the agent image (closes #345)
operations_listed_to now leaves out every session_only Operation for a Credential that resolves as an Access Key, read-only or not, beside the read-only rule, so the listing and Core::execute read the same declarations. A Session, a Runner Key, a Job Key and a call with no Secret are listed as before. Calling one of the eight anyway answers the same refusal. Closes #347 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
fix: the MCP door lists an Access Key none of the tools it refuses one (Closes #347)
On an instance with a Runner, the + list's agent group gains "From a video". Its link goes to an agent Job with no recipe, as pasted text does, so Kamosu does not read the page first and no bare recipe named after the video's page lands. The agent creates the recipe with its Job Key, and it lands in the cook's Cookbook as the Job's own. import_by_agent takes the link as `link`, refused where it is no http or https address on the internet, and lists the arrival under `web`. Both Doors grow it. The group's heading now reads "Kamosu's AI agent makes the recipe · a few minutes", the send sheet and the paste hint use the same verb, and every agent row begins with "From", as the quick ones do. Closes #349 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
A video goes straight to the AI agent from the + list (closes #349)
KAMOSU_RUNNER_INSTANCE and KAMOSU_RUNNER_AGENT_SOCKET stand in for --kamosu and --agent-socket, so a Runner beside an agent's container is set up from one settings file and its Compose command is just [runner]. A flag wins over its variable, an empty variable counts as absent, and the checks are the flags' own. A refused address no longer repeats its password in the log. just runner-start keeps every KAMOSU_RUNNER_ line out of the agent's container. Closes #348 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
The Runner takes its address and agent socket from the environment (closes #348)
An Operator adds rules to the end of the agent's instructions from the Operator screen, without mounting a file or building an image. The Runner section gains one row, which opens a sheet to write them. - get_house_notes and set_house_notes, Operator only and never a Job Key, store up to 4,000 characters with who saved them and when (migration 53). - ask_for_work hands each Job the notes as they stand when it is taken, as house_notes. The Runner passes the Job on unread, so the Job Key's reach does not change. - The published agent appends them to its instructions under a heading saying they are the Operator's and win where the two disagree. - ADR 0048 amended, House Notes in the glossary, docs/runner.md documents the field. Closes #344 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
An Operator writes House Notes for the agent (closes #344)
Closes #353. - A pack size after the amount is taken out before the line is read: `2 x 400 g tins of whole peeled tomatoes` reads as `2 tins of whole peeled tomatoes` does. The amount stays the `2` the line wrote, never `800 g`. The sign may be `x`, `X` or `×`, against either number, and a pack size restated after a slash goes with it, so `2 x 400 g / 14 oz` is never an amount of `2 / 14`. - The `de` (or `d'`) that is a name's second word no longer counts toward the six-word limit on a Food's name, so `lait de coco non sucré en conserve` reads whole. This gives one word of room, and trims nothing: a still longer French name reads with no Food, and a seven-word sentence opening `X de` now reads as one. - A comma straight after an opening aside in brackets ends nothing: `(white/lower part), finely sliced Spring onion` reads as `finely sliced Spring onion`. A line that opens with a bare comma is unchanged. Across the 86-recipe corpus, the bracket line is the only Reading that changes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
fix: three shapes of Ingredient Line read with a Food (closes #353)
From an older Version's page, whoever writes the recipe can put that Version back as the recipe, as a new Version on top that never collapses, or start a named variation from it. Anyone else who sees it can keep it as their own copy, named where they already hold one. Tiles at the foot of the page open a sheet each (choice C, with confirmation). Both Doors gain put_back_version and keep_version_as_copy; start_variation takes a version_id; changed_since answers keeping_needs_name. Closes #354 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
Put an older Version back, or start a variation from it (closes #354)
A save matched old and new Ingredient Lines by position, so inserting, removing or reordering a line read every line below it again. That lost set_reading corrections and could name new Foods after the misreads. kept_lines now pairs each new line with the old line it keeps word for word, wherever it sits, the same words twice matched in order. The carry and the set of lines to read are both taken from that one mapping. A line whose words changed at all is still read again. The MCP Door's instructions and the Catalogue no longer say a line keeps its Reading only "in the same place". Closes #355 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…ding fix: a line that only moved keeps its Reading on save
When a Translation is made, or a save brings it up to a newer source Version, each Ingredient Line takes the Food of the source line in its place, where both lists have the same shape and the two lines read the same amount and Unit. Only the Food is taken: the Reading keeps the line's own words and no Food learns a name. A plain edit reads the lines it wrote on their own words. Recorded as ADR 0050. Closes #356 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…s-source-foods A Translation's line takes the Food of the line it renders (closes #356)
#356 lengthened start_translation's description until the MCP listing had room for all but one of its answer's field meanings, under Claude Code's 2,048-character cap, and the parity test failed on dev. Two sentences that restated "an ordinary Branch" go, and the instruction #356 added is said in fewer words. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
"Ladle the curry" in a Vietnamese beef curry linked the curry powder, since "curry" is a main word of that name. A mention that is only part of an ingredient's name, and that the recipe's title also holds, now names the dish and links nothing. A whole name still links, and a part the title does not hold is unaffected. step_uses takes the title from the one caller, which already held the content. Closes #357 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…n wins Review of #357 found two misses. A title's "de" matched a Step's "dés", so a French title dropped the link to "dés de jambon"; the title is now compared with the accent counting, as a name made of such words is. And a part naming the dish was dropped before it could beat a shorter mention of another line it overlaps, which then linked; it now claims its words first and links nothing after. The changelog names the commoner loss, "the chicken" in a roast chicken. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
fix: a Step naming the dish by its title links no ingredient (Closes #357)
create_recipe, save_recipe_version, edit_recipe and start_translation answer foods_created: each Food reading the save's lines created, with its id, the word, its Language and the lines read as it. A Food the save only found is not listed, nor one an earlier save created, so a collapsed save lists its own and an unchanged one lists none. read_unread_lines now says which Foods it created instead of only a count. The four answers' field meanings must still fit the MCP listing's 2,048-character cap, so the summaries of save_recipe_version and start_translation lose a few restated phrases. The promotions answer as they declare, without the new field. Closes #358 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
…-created feat: a save names the Foods it created (Closes #358)
Sets the version to 0.3.0 in Cargo.toml, Cargo.lock, ui/package.json, ui/package-lock.json, and dates the changelog's Unreleased section as v0.3.0. Made by `just release-prepare 0.3.0`.
The README, the Compose file and both Docker Hub pages now name 0.3.0 and describe what shipped: a page read from its own text, the Runner's imports, House Notes, putting an older version back, and the files kept for an agent. ADR 0048 is amended where #279, #317, #318, #325 and #349 went past it, and the glossary gains Agent and Read Again and brings Operator, Runner and Job level. The agent image's rebuild line moves to 2026-10-18 for Alpine's security fixes to libx11 and libxrender published that day. rustls moves to 0.23.46 for RUSTSEC-2026-0346 to 0349. The three zip advisories are in `extract`, which no shipped code calls, and are recorded in deny.toml. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Stsb9AADG6iuPsFKbsNCgy
Prepare the v0.3.0 release
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
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.
Releases v0.3.0: merges
devintomain.The release was reviewed on the version-bump pull request, #366.
just release-publish 0.3.0merges this pull request, and refuses while any box in that pull request's checklist is unticked. Publishing waits for the maintainer to approve the release title and notes.