Repository navigation
feat: your skill and knowledge review requests in Claude Code — /altimate-code:rules, decision toasts, publish relays the review - #22
Conversation
…d in /altimate-code:rules, decisions as toasts, publish relays the review (1.6.0) - `/altimate-code:rules` also lists the person's skill and knowledge requests to the linked workspace (`workspace review status --json`): status, update, reviewer, reason, waiting over 14 days; replaced ones counted, not listed. Read-only: withdraw and remove happen in the web app, altimate-code's TUI or VS Code extension. - `hooks/review-notices.ts`: in an interactive session, at start and every 15 minutes, asks altimate-code for decisions this host has not toasted (`workspace review decisions --json --unseen --mark-seen --host claude-ui --directory <cwd>`) and toasts each once (three at most per check, the rest counted). A CLI without the command, or a project that is not linked or signed in, stops the checks for the session. - The skill's publish guidance uses `skill publish --json` and relays `review` (sent for review to the owner, already waiting, already approved) or the refusal. - Tests: the toast module (wording, overflow, the CLI call, re-asking, stopping, -p sessions); the resolver check counts the new block. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
| } catch (error) { | ||
| // No CLI at all, or it did not answer in time: try again next time. | ||
| $.ui.log(`review-notices · check failed: ${String(error)}`, { to: 'debug' }) | ||
| return !/not found|ENOENT/i.test(String(error)) |
There was a problem hiding this comment.
[SUGGESTION]: Reuse MISSING_BINARY instead of a second, narrower regex
altimate-cli.ts already exports MISSING_BINARY for exactly this decision (integrations.tsx:114 uses it), and runAltimate only rethrows an error when that regex did not match, after trying both binary names. A start failure whose message says "no such file", "cannot start" or "failed to spawn" passes MISSING_BINARY, is rethrown here, but does not match this narrower /not found|ENOENT/, so checkOnce returns true and the hook keeps spawning the CLI every 15 minutes for the rest of the session. Using the shared regex keeps both places in agreement; this catch branch also has no test.
There was a problem hiding this comment.
Fixed in aa2e348: the catch uses MISSING_BINARY from altimate-cli.ts. Test: a "failed to spawn" error ends the checks, and a timeout keeps them.
| if (await checkOnce($, cwd)) $.clock.after(POLL_MS, () => void tick().catch(() => {})) | ||
| } | ||
| // Off the start's path: the first check runs once the session is up. | ||
| $.clock.after(0, () => void tick().catch(() => {})) |
There was a problem hiding this comment.
[SUGGESTION]: One unexpected failure silently ends the checks for the whole session
tick() reschedules only after checkOnce resolves, and both .catch(() => {}) handlers swallow everything: a single throw ($.ui.toast on a session that just ended, $.clock.after on a dead engine) stops the 15-minute checks for the rest of the session with no log line. later() in altimate-cli.ts:41 already exists for "schedule this off the caller's path and log failures"; using it, and rescheduling from a path that runs even when a check fails, keeps the checks alive and debuggable. startReviewNotices also keeps no handle, so a second session.start on the same engine (resume, /clear) starts a second chain — lineage.tsx cancels its per-session work on session.end, which this loop has no equivalent for.
There was a problem hiding this comment.
Fixed in aa2e348: tick runs through later(). A check that throws is logged, and the next one is still scheduled. A second startReviewNotices (resume, /clear) replaces the earlier chain, through a generation counter, instead of running beside it.
| [ -n "$ALTIMATE_BIN" ] || { echo "altimate-code CLI not found (tried altimate, altimate-code)" >&2; exit 127; } | ||
| "$ALTIMATE_BIN" workspace review status --json < /dev/null | ||
| ``` | ||
| - Exit 0 answers JSON with `workspace_name`, `items`, `skipped` and `complete`. Each item has `kind` (`skill` or `knowledge`), `name`, `status` (`pending`, `active`, `rejected`, `superseded`, `withdrawn` or `retired`), `change` (`add` or `edit`), `reviewed_by_name`, `review_reason` and `expired`. |
There was a problem hiding this comment.
[SUGGESTION]: complete and skipped are named but never used
The contract says exit 0 answers workspace_name, items, skipped and complete, but the workflow only describes items, so a partial answer (complete: false, or a non-empty skipped) would be presented to the user as the whole list of requests. Say what to do with them, e.g. "when complete is false, say some requests may be missing". Step 1 now produces two different JSON documents and step 2 ("The JSON has workspace …") reads as if it described this one; naming each answer (proposals vs review status) would keep the model from mixing them.
There was a problem hiding this comment.
Fixed in aa2e348: a partial list (complete false or skipped above 0) is now flagged with a pointer to the Reviews page. Step 2 now names the proposals JSON, so it isn't confused with the review status.
| /** The host name altimate-code keeps this toast's seen state under: apart from the session-start context's. */ | ||
| export const HOST = 'claude-ui' | ||
| /** How often a running session asks again: decisions arrive while people work. */ | ||
| export const POLL_MS = 15 * 60_000 |
There was a problem hiding this comment.
[SUGGESTION]: The new periodic backend call is not reflected in PRIVACY.md
This hook asks the Altimate backend at every interactive session start and then every 15 minutes (POLL_MS) through workspace review decisions, and shows reviewer names and reasons. PRIVACY.md keeps a dated log of what the plugin sends and still reads "Last revised: November 10, 2026" with the plugin 1.5.0 entry, so this release's new requests are undocumented; add a 1.6.0 entry and bump the revised date.
There was a problem hiding this comment.
Fixed in aa2e348: PRIVACY.md has a "Your skill and knowledge requests" section (1.6.0) covering the checks, what is read and where it is kept. The revised date is now November 13, 2026.
Code Review SummaryStatus: 2 Issues Found | Recommendation: Address before merge Overview
All four findings from the previous review are fixed at this head: the catch reuses Issue Details (click to expand)SUGGESTION
Files Reviewed (11 files)
Fix these issues in Kilo Cloud Previous Review Summary (commit 601e70d)Current summary above is authoritative. Previous snapshots are kept for context only. Previous review (commit 601e70d)Status: 4 Issues Found | Recommendation: Address before merge Overview
The core logic is sound: Issue Details (click to expand)SUGGESTION
Files Reviewed (10 files)
Reviewed by deepseek-v4.1-flash-1017 · Input: 126.9K · Output: 24.4K · Cached: 1.9M |
…is cut - `workspace review decisions --unseen --claim` holds the decisions for this session; once toasted, `--ack <key>…` marks them toasted. Two sessions never toast one decision, and one whose session ended before its toast is shown again after the hold. - Toasts cut long names (80), reviewers (60) and reasons (160). - CHANGELOG: the first check records kept decisions without a toast; Claude is told of decisions whether or not team rules apply. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Kilo Code Review could not run — your account is out of credits. Add credits or switch to a free model to enable reviews on this change. |
ralphstodomingo
left a comment
There was a problem hiding this comment.
Read at 5ea62478b0. I didn't run it. The calls match AltimateAI/altimate-code#1716 at fd262ad200:
workspace review decisionstakes--claim,--ack,--hostand--directory. It always answers JSON and exits 0, and its fields and final statuses are the onesreview-notices.tsreads.workspace review status --jsongives what/altimate-code:ruleslists.
There's one thing to fix before merge, and a version question.
Fix before merge: skill publish --json on a released CLI (inline). The altimate-code skill now runs skill publish "<name>" --json. The latest release, 0.12.10, has no --json on skill publish, and its parser rejects unknown arguments, so publishing from Claude fails for everyone on it. The marketplace serves main, so this ships on merge, and nothing here checks the CLI version. Even merged after the next altimate-code release, it fails for anyone who hasn't updated. Telling Claude to run it again without --json when the CLI says the argument is unknown would keep publishing working on an older CLI.
Version: 1.6.0 is also planned for #21 (column impact). The two conflict in CHANGELOG.md and in the imports of hooks/integrations.tsx; both conflicts are trivial. Each merge to main is a release, so whichever merges second takes 1.7.0. Which do you expect to ship first?
Checked and fine:
- Older CLIs. Without
workspace review decisions, the toast gets no JSON and stops for the session. Withoutworkspace review status,/altimate-code:rulesleaves the section out. - The toast's claim and ack. A decision is marked shown only after its toast, and a lapsed claim hands it out again. Each check shows at most 3 toasts, the last counting the rest.
- No conflict with #16: a trial merge is clean.
Smaller:
/altimate-code:rulesshows names, reviewers and reasons written by other people. The Codex plugin (AltimateAI/altimate-codex-plugin#8) tells the model to show them as quoted data and not to follow instructions in them. The same line here would keep the two plugins alike.- The description still says
--mark-seen; the code uses--claimand--ack.
Appendix — complexity delta (review requests in Claude Code)
15ef117ef6 → 5ea62478b0 · only functions this diff touches · advisory, not a gate.
✅ No touched function changed in complexity (1 touched, 13 new, all under 10).
ℹ️ How to read these numbers
Cognitive (Sonar spec) counts breaks in linear reading flow — each if/loop/catch/ternary/boolean-operator switch adds 1, and nesting makes every further break cost more. It approximates how much you must hold in your head to follow the function: 0–5 trivial · 6–10 easy · 11–15 moderate (15 = Sonar's recommended per-function cap) · 16–25 hard to follow · >25 needs decomposition.
CCN (cyclomatic) counts independent paths — also the minimum number of test cases for full branch coverage of the function.
Only functions this diff touches are measured, as deltas — pre-existing complexity is not counted against this change. Rising numbers aren't automatically wrong; they're where review attention should go. Test files excluded.
| - **Check the link:** `"$ALTIMATE_BIN" workspace status < /dev/null`. Exit code 0 means linked and 3 means not linked. Exit code 2 with `Not signed in to Altimate` means the user must sign in. For anything else, show the user the CLI's message. | ||
| - **Link:** do not suggest `! altimate link`: its picker needs a terminal, and this session does not give it one. Once the user chooses a workspace (from `workspace_context`, or `"$ALTIMATE_BIN" workspace list < /dev/null`), run `ALTIMATE_DISABLE_MEMORY=1 "$ALTIMATE_BIN" link --workspace "<id or name>" < /dev/null`. `ALTIMATE_DISABLE_MEMORY=1` stops linking from uploading this machine's existing memory to the workspace, which the user has not agreed to here. For a new workspace, run `ALTIMATE_DISABLE_MEMORY=1 "$ALTIMATE_BIN" link --create < /dev/null`, which names it after the repository. If a workspace with that name already exists, `--create` refuses and names it: offer to link to that one with `--workspace`, or, if the user confirms they want a second one, re-run with `--allow-duplicate`. If the project is already linked to a different workspace, `link` refuses; confirm the move with the user, then re-run it with `--yes`. Never pass `--claude-md` unless the user asks for it. | ||
| - **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill with the team, run `"$ALTIMATE_BIN" skill publish "<name>" < /dev/null` only for a skill the user named; add `--replace` only after they confirm overwriting the workspace copy. Personal skills in `~/.claude/skills` are refused; copy one into the project only if the user asks. | ||
| - **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill with the team, run `"$ALTIMATE_BIN" skill publish "<name>" --json < /dev/null` only for a skill the user named; add `--replace` only after they confirm overwriting the workspace copy. Relay what the JSON says, not what you expect (its `message` already says it in one line): `{"ok": false, "error": …}` is a refusal or failure, shown as it is. With `ok` true, `review` is `null` when the skill was attached directly (no review on that workspace). Otherwise read `review.status` and `review.duplicate`: `pending` means it was sent for review to `review.owner_name` and reaches the workspace once approved (`pending` with `duplicate` true means the same content is already waiting); `active` means it is in the workspace now (`active` with `duplicate` true means that content was already approved). `/altimate-code:rules` shows where the user's skill and knowledge requests stand. Personal skills in `~/.claude/skills` are refused; copy one into the project only if the user asks. |
There was a problem hiding this comment.
0.12.10, the latest altimate-code release, has no --json on skill publish, and its root parser is strict. So this command is refused there as an unknown argument, and the skill is never published. Because the marketplace serves main, this reaches everyone as soon as it merges, whatever CLI they have. One sentence would keep publishing working on an older CLI: if the CLI says --json is an unknown argument, run the same command without it and relay its text.
There was a problem hiding this comment.
Fixed in cc48cd0. Claude first runs skill publish --help, which has no side effects. If it lists --json, Claude publishes with it; otherwise Claude publishes without it, relays the text and suggests npm install -g altimate-code@latest. The publish runs once and is never rerun after a failure.
hkaltai
left a comment
There was a problem hiding this comment.
Peer review: request changes
I reviewed this with independent Claude and Codex (Sol 6.1) reviewers plus a lead pass, and ran one convergence round. Both reviewers agree on the verdict.
Blocking (MAJOR)
- The toast acknowledgement fails every time, and the plugin ignores the failure, so decisions are toasted again on every check. The root cause is in altimate-code#1716 (
ackHostDecisionsin a fresh process). skill publish "<name>" --jsonbreaks publishing on every released altimate-code, because the parser is.strict()and--jsononskill publishis new in #1716.
There are four more minor inline comments. Nits are in a separate comment. The PR description also says --unseen --mark-seen, but the code uses --claim and --ack; please update it.
Missing tests: an ack that answers {ok:false}; publishing against a CLI without --json; rendering when complete: false or skipped > 0; several toasts in one check against the real engine.
| // Shown (the last toast counts the rest): acknowledged, so they are not handed out again. A failed acknowledgement | ||
| // costs a repeat after the hold, never a lost decision. | ||
| const keys = decisions.map(d => d.key).filter((k): k is string => typeof k === 'string') | ||
| if (keys.length) |
There was a problem hiding this comment.
if (keys.length)
MAJOR (bug, cross-repo). The ack runs in a fresh CLI process. In altimate-code#1716, ackHostDecisions gets its scope from RulesSync.knownAccount(), an in-memory map that only a poll fills, so this ack always answers {"ok":false,"status":"failed"}. The plugin never reads the answer and only catches thrown errors. The claim lapses after 5 minutes, and each 15-minute check toasts the same decisions again until they age out.
Fix: fix the root cause in altimate-code (see the comment on #1716). Here, parse the ack's stdout and log at debug level when ok !== true. The stubbed ack in review-notices.test.ts can't catch this. (Claude reviewer; I confirmed it in the code. Codex agreed in convergence.)
There was a problem hiding this comment.
Root cause fixed in AltimateAI/altimate-code#1716 (4466331199): --ack now resolves the account in a process that never polled, with a test. Here (aa2e348), the ack's stdout is parsed and anything but ok: true is logged at debug level, with a test.
| - **Check the link:** `"$ALTIMATE_BIN" workspace status < /dev/null`. Exit code 0 means linked and 3 means not linked. Exit code 2 with `Not signed in to Altimate` means the user must sign in. For anything else, show the user the CLI's message. | ||
| - **Link:** do not suggest `! altimate link`: its picker needs a terminal, and this session does not give it one. Once the user chooses a workspace (from `workspace_context`, or `"$ALTIMATE_BIN" workspace list < /dev/null`), run `ALTIMATE_DISABLE_MEMORY=1 "$ALTIMATE_BIN" link --workspace "<id or name>" < /dev/null`. `ALTIMATE_DISABLE_MEMORY=1` stops linking from uploading this machine's existing memory to the workspace, which the user has not agreed to here. For a new workspace, run `ALTIMATE_DISABLE_MEMORY=1 "$ALTIMATE_BIN" link --create < /dev/null`, which names it after the repository. If a workspace with that name already exists, `--create` refuses and names it: offer to link to that one with `--workspace`, or, if the user confirms they want a second one, re-run with `--allow-duplicate`. If the project is already linked to a different workspace, `link` refuses; confirm the move with the user, then re-run it with `--yes`. Never pass `--claude-md` unless the user asks for it. | ||
| - **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill with the team, run `"$ALTIMATE_BIN" skill publish "<name>" < /dev/null` only for a skill the user named; add `--replace` only after they confirm overwriting the workspace copy. Personal skills in `~/.claude/skills` are refused; copy one into the project only if the user asks. | ||
| - **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill with the team, run `"$ALTIMATE_BIN" skill publish "<name>" --json < /dev/null` only for a skill the user named; add `--replace` only after they confirm overwriting the workspace copy. Relay what the JSON says, not what you expect (its `message` already says it in one line): `{"ok": false, "error": …}` is a refusal or failure, shown as it is. With `ok` true, `review` is `null` when the skill was attached directly (no review on that workspace). Otherwise read `review.status` and `review.duplicate`: `pending` means it was sent for review to `review.owner_name` and reaches the workspace once approved (`pending` with `duplicate` true means the same content is already waiting); `active` means it is in the workspace now (`active` with `duplicate` true means that content was already approved). `/altimate-code:rules` shows where the user's skill and knowledge requests stand. Personal skills in `~/.claude/skills` are refused; copy one into the project only if the user asks. |
There was a problem hiding this comment.
- **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill w...
MAJOR (backward compatibility). --json on skill publish is new in altimate-code#1716, and the CLI parses with .strict(). Every released altimate-code (for example v0.12.8 and v0.12.10) rejects this with Unknown argument: json. Publishing then breaks where it works with the current plugin. That contradicts the PR's claim that "with an older altimate-code, the new parts stay quiet". The instructions here only cover {"ok": false, …} JSON, not this plain-text failure.
Fix: if the output says Unknown argument: json, rerun without --json, relay the text, and suggest npm install -g altimate-code@latest. Retry only on that argument error, never on an ambiguous publish failure. Alternatively, gate this on a minimum CLI version. (Claude rated it MAJOR and Codex MINOR; I kept MAJOR because it breaks a working feature.)
There was a problem hiding this comment.
Fixed in cc48cd0. Claude first runs skill publish --help, which has no side effects. If it lists --json, Claude publishes with it; otherwise Claude publishes without it, relays the text and suggests npm install -g altimate-code@latest. The publish runs once and is never rerun after a failure. The PR description says so too.
| if (!answer || typeof answer.ok !== 'boolean') return false | ||
| if (!answer.ok) return !FINAL.has(answer.status ?? '') | ||
| const decisions = answer.decisions ?? [] | ||
| for (const text of toastsFor(decisions)) $.ui.toast(text, { timeoutMs: 10_000 }) |
There was a problem hiding this comment.
for (const text of toastsFor(decisions)) $.ui.toast(text, { timeoutMs: 10_000 })
MINOR (plausible). Consecutive fire-and-forget $.ui.toast calls, followed at once by acknowledging every key, can mark decisions seen that were never readable. The engine's notification bar can replace a plugin's earlier toast with its newer one. Once the ack works, that becomes silent loss.
Fix: show one combined toast for the decisions and the overflow count, or acknowledge only after display. (Codex reviewer; not checked against a live engine.)
There was a problem hiding this comment.
Fixed in aa2e348: each check shows one toast, the newest decision plus how many more (pointing to /altimate-code:rules). Everything a check acknowledges was said in that one toast. The engine docs confirm a plugin's newer toast takes its own earlier one's place on the notification bar, so this was a real risk.
| /** Toasts per check; the rest are counted in the last one. */ | ||
| export const MAX_TOASTS = 3 | ||
| /** Answers that mean nothing will change in this session: stop asking. */ | ||
| const FINAL = new Set(['signed_out', 'not_linked', 'unsupported', 'off']) |
There was a problem hiding this comment.
const FINAL = new Set(['signed_out', 'not_linked', 'unsupported', 'off'])
MINOR (logic). For any binding outcome other than bound, including unknown (the service couldn't be reached), the CLI answers not_linked. The plugin treats that as final, so a brief network failure at session start turns toasts off for the whole session. A sign-in during the session also gets no toasts until the next one.
Fix: keep retrying, with backoff, on not_linked and signed_out, or have the CLI map unknown to failed. (Claude reviewer; Codex agreed on the binding-failure case.)
There was a problem hiding this comment.
Fixed on both sides. In AltimateAI/altimate-code#1716 (4466331199), a link that can't be confirmed is now failed, not not_linked. Here (aa2e348), signed_out and not_linked are no longer final: the next 15-minute check asks again, which also covers a sign-in or link during the session. Only unsupported and off stop it.
| 2. **Show the state.** The JSON has `workspace` (`name`, `owner_name`, `is_owner`; `null` when the project is not linked), `proposals`, `team_rules` and `mine`. Show, briefly: | ||
| - the number of team rules in the workspace, and their texts when there are 10 or fewer; | ||
| - the user's own proposals in `mine` that are not yet active, with their `status` and, for a rejected one, the `reason`; | ||
| - the user's skill and knowledge requests, newest first: kind, name, status (`pending` → waiting for review, `active` → approved, `rejected`, `withdrawn`, `retired` → removed from the workspace), "update" for `change: edit`, the reviewer and the reason when given, and "waiting over 14 days" when `expired`. Do not list `superseded` ones (replaced by a newer request); say how many there were instead. Withdrawing a waiting request or removing an approved one happens in the Altimate web app, the altimate-code TUI (`/rules`) or the VS Code extension, not here; |
There was a problem hiding this comment.
- the user's skill and knowledge requests, newest first: kind, name, status (`pending` → waiting for review, `active` → approved, `rejected`, `withdrawn`, `retired` → removed from the workspace), "update" for `change:...
MINOR (API contract). Two gaps in how the listing is shown:
completeandskippedare named as fields but never shown. The CLI caps the listing and skips rows it can't parse, so a partial list reads as complete. The CLI's text form warns about both (workspace.ts:360-362).- "waiting over 14 days" applies whenever
expiredis set. The CLI shows it only whenstatus === "pending".
Fix: warn when complete is false or skipped > 0, and point to the workspace's Reviews page. Show the 14-day marker only for pending. (Codex and Claude reviewers.)
There was a problem hiding this comment.
Fixed in aa2e348: a partial list is flagged and points to the workspace's Reviews page, and "waiting over 14 days" is shown only for a pending request with expired.
|
Peer review: nits (non-blocking)
I reviewed this with independent Claude and Codex (Sol 6.1) reviewers. |
…publish on an older CLI - one toast per check (the newest decision and how many more), so every decision a check acknowledges was said - the `--ack` answer is read and a refusal is logged; a check that throws is logged and the next is still scheduled; a second `session.start` replaces the earlier check chain - signed out and not linked are asked again at the next check; `MISSING_BINARY` decides when there is no CLI - `skill publish`: on an altimate-code without `--json`, Claude reruns it once without it and relays the text - `/altimate-code:rules`: says when the request list may be partial, shows "waiting over 14 days" only for pending requests, and treats names, reviewers and reasons as quoted data - PRIVACY.md: the review status checks Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
hkaltai
left a comment
There was a problem hiding this comment.
Re-review: request changes
I re-reviewed the whole PR at aa2e348 with independent Claude and Codex (Sol 6.1) reviewers plus a lead pass, then ran one round where both reviewers checked the combined findings. Both reviewers agree on the verdict.
Fixed:
- The ack answer is now read and logged, and the CLI resolves the account in a fresh process.
- One toast per check.
complete/skippedwarnings.- Expiry shown for pending requests only.
not_linkedandsigned_outno longer stop the checks.- The PR description now matches the code.
MISSING_BINARYis reused.
Still blocking (MAJOR): the publish fallback for older CLIs never fires. Details are inline. I reproduced it on altimate-code 0.12.8.
Nit: the test name at hooks/review-notices.test.ts:114 ("…or a project that is not linked, is not asked again") no longer matches the behaviour. Since the fix, an unlinked project is asked again. Rename it to "a CLI without the command is not asked again".
| - **Check the link:** `"$ALTIMATE_BIN" workspace status < /dev/null`. Exit code 0 means linked and 3 means not linked. Exit code 2 with `Not signed in to Altimate` means the user must sign in. For anything else, show the user the CLI's message. | ||
| - **Link:** do not suggest `! altimate link`: its picker needs a terminal, and this session does not give it one. Once the user chooses a workspace (from `workspace_context`, or `"$ALTIMATE_BIN" workspace list < /dev/null`), run `ALTIMATE_DISABLE_MEMORY=1 "$ALTIMATE_BIN" link --workspace "<id or name>" < /dev/null`. `ALTIMATE_DISABLE_MEMORY=1` stops linking from uploading this machine's existing memory to the workspace, which the user has not agreed to here. For a new workspace, run `ALTIMATE_DISABLE_MEMORY=1 "$ALTIMATE_BIN" link --create < /dev/null`, which names it after the repository. If a workspace with that name already exists, `--create` refuses and names it: offer to link to that one with `--workspace`, or, if the user confirms they want a second one, re-run with `--allow-duplicate`. If the project is already linked to a different workspace, `link` refuses; confirm the move with the user, then re-run it with `--yes`. Never pass `--claude-md` unless the user asks for it. | ||
| - **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill with the team, run `"$ALTIMATE_BIN" skill publish "<name>" < /dev/null` only for a skill the user named; add `--replace` only after they confirm overwriting the workspace copy. Personal skills in `~/.claude/skills` are refused; copy one into the project only if the user asks. | ||
| - **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill with the team, run `"$ALTIMATE_BIN" skill publish "<name>" --json < /dev/null` only for a skill the user named; add `--replace` only after they confirm overwriting the workspace copy. If the output instead says `json` is an unknown argument (`Unknown argument: json`), the CLI predates `--json` on `skill publish`: run the same command once more without `--json`, relay its text as it is, and suggest updating altimate-code (`npm install -g altimate-code@latest`). Rerun only for that argument error, never after any other failure, which may have published. Relay what the JSON says, not what you expect (its `message` already says it in one line): `{"ok": false, "error": …}` is a refusal or failure, shown as it is. With `ok` true, `review` is `null` when the skill was attached directly (no review on that workspace). Otherwise read `review.status` and `review.duplicate`: `pending` means it was sent for review to `review.owner_name` and reaches the workspace once approved (`pending` with `duplicate` true means the same content is already waiting); `active` means it is in the workspace now (`active` with `duplicate` true means that content was already approved). `/altimate-code:rules` shows where the user's skill and knowledge requests stand. Personal skills in `~/.claude/skills` are refused; copy one into the project only if the user asks. |
There was a problem hiding this comment.
- **Skills:** `"$ALTIMATE_BIN" skill list < /dev/null` lists them; `"$ALTIMATE_BIN" workspace refresh < /dev/null` pulls the latest workspace skills (exit 3 means not linked: offer to link). To share a project skill w...
MAJOR (backward compatibility, still open). The fallback reruns without --json only when the output says Unknown argument: json, but released altimate-code never prints that. Its custom yargs .fail handler (packages/opencode/src/index.ts:282-293, the same in v0.12.10) calls cli.showHelp() and then process.exit(1) without printing msg.
I reproduced it on altimate-code 0.12.8: altimate-code skill publish <name> --json prints the skill publish <name> help and exits 1, with no "Unknown argument" text. The skill also says to rerun only on that error, so Claude reports a failure, and publishing from a session still breaks on every released CLI.
Fix: probe before publishing, which has no side effects. If "$ALTIMATE_BIN" skill publish --help < /dev/null lists --json, use it. Otherwise publish without it, relay the text, and suggest npm install -g altimate-code@latest. If you'd rather not probe, the weaker option is to treat "exit 1 with the usage/help screen and no JSON" as the trigger. (Claude found this; I reproduced it; Codex agreed in convergence.)
There was a problem hiding this comment.
Confirmed: the .fail handler shows help and exits 1 without the message, and v0.12.10's skill publish has no --json. Fixed in cc48cd0. Claude first runs skill publish --help, which has no side effects. If it lists --json, Claude publishes with it; otherwise Claude publishes without it, relays the text and suggests npm install -g altimate-code@latest. The publish runs once and is never rerun after a failure. On this branch's CLI, skill publish --help exits 0 and lists --json Print the result as JSON. The test name is fixed in the same commit.
Released altimate-code answers an unknown option with its help screen and no message, so the `Unknown argument` fallback never fired. Claude now reads `skill publish --help` (no side effects) and publishes with `--json` only when it is listed. The review-notices test name matches the behaviour: an unlinked project is asked again. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks @ralphstodomingo and @hkaltai, and for the re-review. Both blocking items are fixed in aa2e348 and cc48cd0, with the CLI side in AltimateAI/altimate-code#1716 (4466331199). Each inline thread has a reply. Blocking:
Smaller:
Version: the merge order with #21 isn't settled yet. Whichever merges second moves to 1.7.0 and resolves the CHANGELOG and import conflicts. Missing tests:
Could you take another look? |
hkaltai
left a comment
There was a problem hiding this comment.
Re-review: approve
I re-reviewed the whole PR at cc48cd0 with independent Claude and Codex (Sol 6.1) reviewers plus a lead pass, then ran one round where both reviewers checked the combined findings. Both approve.
The previous blocker is fixed. SKILL.md:238 now probes skill publish --help before publishing, which changes nothing. It uses --json only when the help lists it, publishes once, and never reruns after a failure. Both reviewers traced old and new CLIs in the source: v0.12.9 has only --replace and no global *json* option, while #1716 registers --json. I also checked the installed altimate-code 0.12.8, where skill publish --help exits 0 and doesn't list --json, so the skill takes the plain-text path. All the other earlier findings are fixed too, including the test name.
This was a source review plus that one help-output check. I didn't run the plugin tests or a live Claude Code session.
Two nits, neither blocking:
hooks/review-notices.test.ts:115: the old-CLI stub printsUnknown argument: …on stdout. A released CLI prints help on stderr, leaves stdout empty and exits 1. The behaviour is the same, since neither is JSON, but{ exitCode: 1, stdout: '' }would match reality. That same mistaken assumption caused the last round's blocker.skills/altimate-code/SKILL.md:238: optionally note that the help is printed on stderr, so anyone porting this to a script that reads only stdout doesn't always take the plain-text branch.
Remember to merge and release this after the altimate-code release that includes #1716.
ralphstodomingo
left a comment
There was a problem hiding this comment.
Re-review at cc48cd07bc. Approving: the publish point is fixed.
- Publishing on an older CLI. Claude now runs
skill publish --helpfirst and adds--jsononly when the help lists it. I checked against 0.12.10: itsskill publishtakes only--replace, and there's no global--json. So an older CLI publishes without--json, and the publish still runs only once. - The smaller points.
/altimate-code:rulesnow shows names, reviewers and reasons as quoted data, and the description describes--claimand--ack.
On the version, it's fine to settle the order at merge time. #23 now claims 1.6.0 too, alongside #21. Of the three, whichever merges second takes 1.7.0, and the third takes 1.8.0.
Appendix — complexity delta (review requests in Claude Code)
15ef117ef6 → cc48cd07bc · only functions this diff touches · advisory, not a gate.
| Function | File | Cognitive | CCN | Status |
|---|---|---|---|---|
checkOnce L69 |
plugins/altimate-code/hooks/review-notices.ts |
new → 7 | new → 10 | new — easy |
Summary: 1 touched · 1 rose · 0 improved · 12 new (max cognitive 7) · net cognitive Δ +21
ℹ️ How to read these numbers
Cognitive (Sonar spec) counts breaks in linear reading flow — each if/loop/catch/ternary/boolean-operator switch adds 1, and nesting makes every further break cost more. It approximates how much you must hold in your head to follow the function: 0–5 trivial · 6–10 easy · 11–15 moderate (15 = Sonar's recommended per-function cap) · 16–25 hard to follow · >25 needs decomposition.
CCN (cyclomatic) counts independent paths — also the minimum number of test cases for full branch coverage of the function.
Only functions this diff touches are measured, as deltas — pre-existing complexity is not counted against this change. Rising numbers aren't automatically wrong; they're where review attention should go. Test files excluded.
|
Heads-up: plugin 1.6.0 just shipped from #21 (column impact), so this one needs the next version: 1.7.0 in |
main released 1.6.0 (column impact) and 1.6.1 while this branch also claimed 1.6.0, so this branch's entry becomes 1.7.0: - `plugin.json` and `marketplace.json` at 1.7.0; the changelog entry is `[1.7.0] — 2026-11-17`, above main's 1.6.1 and 1.6.0 - PRIVACY.md: the review status checks are "since plugin 1.7.0" - `hooks/integrations.tsx`: main's imports (`task-impact` replaced `lineageBand`) plus this branch's `review-notices` Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
|
||
| **Effective date:** November 2, 2026 | ||
| **Last revised:** November 10, 2026 | ||
| **Last revised:** November 13, 2026 |
There was a problem hiding this comment.
[SUGGESTION]: "Last revised" now precedes the document's own latest change entry
The same policy's "Changes to this policy" log (line 154, added with the 1.6.0 lineage-report release) is dated November 16, 2026, so setting "Last revised" to November 13, 2026 here makes the revision date older than a change the document already records, and the new 1.7.0 section added by this PR still has no dated entry in that log. Use the release date (2026-11-17, as in CHANGELOG.md) and add a dated line for "Your skill and knowledge requests" (plugin 1.7.0).
| - **"Remember this as a team rule: …"**, said to Claude, waits as a proposal the same way. It needs "as a rule" or "team rule" followed by ":" or "-": "remember to run the tests", a question, or "don't remember this as a team rule" is just part of the conversation. A task handed to altimate-code decides this as an altimate-code session does. | ||
| - **`/altimate-code:rules`** shows the proposals waiting for you, the team rules, and the status of the rules you proposed. For each proposal you choose: send it to the workspace owner's review (or approve it, if you are the owner), edit the wording, save it as a memory, or "Not a rule". Nothing reaches teammates until the owner approves it. | ||
| - **`/altimate-code:rules`** shows the proposals waiting for you, the team rules, and the status of the rules you proposed. For each proposal you choose: send it to the workspace owner's review (or approve it, if you are the owner), edit the wording, save it as a memory, or "Not a rule". Nothing reaches teammates until the owner approves it. It also lists your skill and knowledge requests to the workspace (waiting, approved, rejected with the reason, or removed), read-only: withdrawing a request or removing an approved skill or knowledge base happens in the Altimate web app, altimate-code's TUI (`/rules`) or its VS Code extension. | ||
| - **Decisions on your skill and knowledge requests** are told to Claude at session start, and shown to you as a toast once each: at session start and every 15 minutes while an interactive session runs. This needs an altimate-code release with `workspace review decisions`; with an older one there are no toasts. |
There was a problem hiding this comment.
[SUGGESTION]: "shown to you as a toast once each" describes more than the code does
toastFor (hooks/review-notices.ts:59) shows the text of the newest decision only; any others are reported as "N more decisions …" and are then acknowledged as seen, so with several decisions the user never reads the rest in a toast. The CHANGELOG already says "one toast per check (the newest, and how many more)"; this bullet implies every decision gets its own toast text. Say one toast per check, naming the newest and how many more, with /altimate-code:rules listing them.
Summary
Your own skill and knowledge requests to a linked workspace, and the owner's or an admin's decisions on them, now show in Claude Code sessions. This is read-only: withdrawing and removing happen in the Altimate web app, the altimate-code TUI or the VS Code extension. It needs altimate-code with AltimateAI/altimate-code#1716 (stacked on #1708). Merge and release this after the first altimate-code release that includes it. With an older altimate-code, the toast and the requests section stay quiet, and
skill publishruns without--json./altimate-code:rulesadds "Your skill and knowledge requests" after your proposals, fromworkspace review status --json.complete: falseorskipped) says so.altimate-codeskill now runsskill publish "<name>" --jsonand relays the review outcome: "sent for review to ", "already waiting", "already approved", or the refusal message, then points to/altimate-code:rules. Claude first readsskill publish --help, which has no side effects. When it lists no--json(every release so far), Claude publishes without it and relays the text. Such a CLI would answer the option with its help screen, not an error.hooks/review-notices.ts, started fromintegrations.tsxin interactive sessions):workspace review decisions --json --unseen --claim --host claude-ui. That holds the decisions for this session, so another session gets none of them.--ack <key>…marks those decisions shown. A failed acknowledgement is logged, and the decision is handed out again once the 5-minute hold lapses.unsupportedoroff, a CLI without the command, or no CLI at all.signed_outandnot_linkedare asked again at the next check.claude-ui), apart from the session-start context. altimate-code adds the decisions and the waiting count to that context with no change here.Verification
/altimate-code:rules, and ask Claude to publish a project skill. Have the owner approve it in the web app, then wait for the toast or start a new session. The same decision isn't toasted again.🤖 Generated with Claude Code