Repository navigation
Refresh competitor comparisons and docs from current sources - #9
Conversation
Re-verified all 10 comparison entries against official pricing/docs: plan renames (CodeRabbit Essentials/Team/Advanced), credit pricing (Qodo, Greptile, Macroscope), new capabilities (Cubic diagrams + ticketing, SonarQube/Gitar memory, Copilot ensemble + memory, Semgrep Multimodal BYOK, Codacy Jira tickets). Stale claims removed; every entry re-dated. Docs rewritten against codebase truth: real rules implementation, correct rate limits (30/hr repo, 200/day org, 10/hr force), hosted auto-review default off / self-hosted on, memory threshold defaults, settings paths, diagram gates, comment formats. Fixes three verified bugs: MEMORY_DEFAULTS mirrored stale backend thresholds, ratelimit.go comments listed wrong limits, auto_run toggle showed "on by default" while hosted treats unset as off.
There was a problem hiding this comment.
7 issues found across 11 files
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="web/src/components/marketing/docs-content.tsx">
<violation number="1" location="web/src/components/marketing/docs-content.tsx:2449">
P2: Auto-resolve invokes the LLM-backed `AddressedJudge` before resolving each proximity candidate, so this description understates both behavior and token spend. Document the judge confirmation and its cost instead of promising a pure-diff, free operation.</violation>
</file>
<file name="web/src/app/docs/layout.tsx">
<violation number="1" location="web/src/app/docs/layout.tsx:20">
P2: The new `Overview` link points to `/docs`, but the new docs tree (`web/src/app/docs/`) has no `page.tsx` at that path — only `layout.tsx`. The `/docs` URL is still owned by the legacy marketing page `web/src/app/(marketing)/docs/page.tsx`, so clicking Overview leaves the new docs shell and lands on the marketing full-reference page rendered without this sidebar (and without `DocsPageViewedProbe`). Either add `web/src/app/docs/page.tsx` so Overview renders inside this layout, or the link is pointing users at an inconsistent standalone page that contradicts the new nav structure.</violation>
</file>
<file name="web/src/app/(dashboard)/settings/page.tsx">
<violation number="1" location="web/src/app/(dashboard)/settings/page.tsx:922">
P3: On self-hosted installs the Auto-review toggle now renders OFF by default even though reviews still run unconditionally. The new description says "self-hosted defaults to on", but the toggle's displayed value comes only from the stored setting or AUTO_RUN_TOGGLE.defaultValue (line 1465-1467 / 1915-1919), and the single defaultValue=false cannot reflect deployment type. Backend decideAutoRun/AutoRun short-circuits self-hosted to ALWAYS review regardless of stored settings (backend/internal/pipeline/persona.go), so a fresh self-hosted org shows "Auto-review every PR" off in the settings page while the backend keeps auto-running reviews — contradicting the new description. Previously defaultValue=true at least displayed self-hosted's actual behavior. Consider deriving the shown default from a self-hosted flag when the backend exposes one, or narrow the description wording.</violation>
</file>
<file name="web/src/app/docs/features/memory-tuning/page.tsx">
<violation number="1" location="web/src/app/docs/features/memory-tuning/page.tsx:50">
P2: The new Thresholds paragraph claims "An explicit 0 disables that gate entirely," but that is false for scenario_dedupe. The backend clamps a non-positive dedupe threshold to the 0.95 default (backend/internal/pipeline/scenarios.go scenarioDedupeThreshold), because similarity >= 0 is always true and a 0 would otherwise suppress every new scenario — the opposite of disabling the gate. Qualify the statement so a user setting scenario_dedupe to 0 is not misled into expecting dedup off.</violation>
</file>
<file name="web/public/docs.md">
<violation number="1" location="web/public/docs.md:245">
P3: The claim that auto-resolve involves "no LLM call" contradicts the codebase: the auto-resolve path runs an addressed-judge LLM pass that consumes model tokens (`backend/internal/pipeline/auto_resolve_tokens_test.go` records judge prompt/model/cost tokens), and README/CONTEXT both state "stale threads auto-resolve only after LLM judge confirms the change addressed the finding; line proximity alone never resolves" and "Proximity is prefilter only; addressed-judge must confirm fix." So a push that merely changes the anchored lines does not itself close the thread — a judge call is involved and the thread stays open on judge failure. Please reconcile the wording (drop "no LLM call" and the implication that the push alone resolves) with the actual `autoResolveOnSynchronize` behavior.</violation>
</file>
<file name="web/src/app/docs/faq/page.tsx">
<violation number="1" location="web/src/app/docs/faq/page.tsx:68">
P2: The FAQ now states checkbox-triggered reviews run under a 10/hour per-repo force bucket, but backend/internal/api/handlers_webhook.go:424-428 still documents the same dispatch flow as "the tighter `force=true` rate-limit path (3/hr/repo)". Verify the actual admission bucket size before rollout: if the limit is still 3/hr, these docs (and the matching claims in docs-content.tsx) repeat the misstatement this PR set out to fix; if it is 10/hr, the backend comment is stale and should be corrected alongside.</violation>
</file>
<file name="web/src/lib/pseo/competitors.ts">
<violation number="1" location="web/src/lib/pseo/competitors.ts:174">
P3: Greptile's `ticketing` flag was flipped to true while the same entry's strengths and summary say ticket handling is context-only: "Reviews code against linked Jira tickets, Linear issues, and Confluence pages (read-only — no ticket creation)" and "doesn't ... create tickets." Every other entry sets `ticketing: true` only where ticket creation exists (Cubic "creates Jira/Linear tickets", Codacy "creates tickets from findings", SonarQube/Gitar "creates follow-up issues"), while Greptile explicitly does not create tickets. If the flag means ticket creation/integration uniformly, this contradicts the copy; if it now means context-only, its semantics differ from the other entries. Align the flag with the written behavior or the other entries.</violation>
</file>
Reply with feedback, questions, or to request a fix.
Re-trigger cubic
| }, | ||
| { | ||
| label: "Auto-resolve", | ||
| desc: "On every push, checks whether the new commits changed lines Argus previously flagged (within \u00B13 lines) and resolves those threads. Pure-diff — no LLM call, no token cost — and runs even when auto-review is off.", |
There was a problem hiding this comment.
P2: Auto-resolve invokes the LLM-backed AddressedJudge before resolving each proximity candidate, so this description understates both behavior and token spend. Document the judge confirmation and its cost instead of promising a pure-diff, free operation.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/src/components/marketing/docs-content.tsx, line 2449:
<comment>Auto-resolve invokes the LLM-backed `AddressedJudge` before resolving each proximity candidate, so this description understates both behavior and token spend. Document the judge confirmation and its cost instead of promising a pure-diff, free operation.</comment>
<file context>
@@ -2412,6 +2442,11 @@ export function DocsContent() {
+ },
+ {
+ label: "Auto-resolve",
+ desc: "On every push, checks whether the new commits changed lines Argus previously flagged (within \u00B13 lines) and resolves those threads. Pure-diff — no LLM call, no token cost — and runs even when auto-review is off.",
status: "on by default",
},
</file context>
| desc: "On every push, checks whether the new commits changed lines Argus previously flagged (within \u00B13 lines) and resolves those threads. Pure-diff — no LLM call, no token cost — and runs even when auto-review is off.", | |
| desc: "On every push, checks whether the new commits changed lines Argus previously flagged (within ±3 lines), then asks an LLM judge to confirm the finding was actually fixed before resolving the thread. Runs even when auto-review is off.", |
| </Link> | ||
| <h2 className="text-xs font-mono text-slate-500 uppercase mb-3 tracking-wider">Docs</h2> | ||
| <nav className="space-y-1"> | ||
| <DocLink href="/docs">Overview</DocLink> |
There was a problem hiding this comment.
P2: The new Overview link points to /docs, but the new docs tree (web/src/app/docs/) has no page.tsx at that path — only layout.tsx. The /docs URL is still owned by the legacy marketing page web/src/app/(marketing)/docs/page.tsx, so clicking Overview leaves the new docs shell and lands on the marketing full-reference page rendered without this sidebar (and without DocsPageViewedProbe). Either add web/src/app/docs/page.tsx so Overview renders inside this layout, or the link is pointing users at an inconsistent standalone page that contradicts the new nav structure.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/src/app/docs/layout.tsx, line 20:
<comment>The new `Overview` link points to `/docs`, but the new docs tree (`web/src/app/docs/`) has no `page.tsx` at that path — only `layout.tsx`. The `/docs` URL is still owned by the legacy marketing page `web/src/app/(marketing)/docs/page.tsx`, so clicking Overview leaves the new docs shell and lands on the marketing full-reference page rendered without this sidebar (and without `DocsPageViewedProbe`). Either add `web/src/app/docs/page.tsx` so Overview renders inside this layout, or the link is pointing users at an inconsistent standalone page that contradicts the new nav structure.</comment>
<file context>
@@ -17,9 +17,18 @@ export default function DocsLayout({ children }: { children: ReactNode }) {
</Link>
<h2 className="text-xs font-mono text-slate-500 uppercase mb-3 tracking-wider">Docs</h2>
<nav className="space-y-1">
+ <DocLink href="/docs">Overview</DocLink>
+
+ <p className="pt-4 pb-1 text-[10px] font-mono uppercase tracking-wider text-slate-600">
</file context>
| more relevant matches). Lower = more permissive (more context, more noise). Scores are raw | ||
| cosine similarity over the configured embedding space — they sit high, so the defaults | ||
| are higher than intuition suggests. An explicit{" "} | ||
| <code className="bg-slate-900 px-1 text-amber">0</code> disables that gate entirely. |
There was a problem hiding this comment.
P2: The new Thresholds paragraph claims "An explicit 0 disables that gate entirely," but that is false for scenario_dedupe. The backend clamps a non-positive dedupe threshold to the 0.95 default (backend/internal/pipeline/scenarios.go scenarioDedupeThreshold), because similarity >= 0 is always true and a 0 would otherwise suppress every new scenario — the opposite of disabling the gate. Qualify the statement so a user setting scenario_dedupe to 0 is not misled into expecting dedup off.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/src/app/docs/features/memory-tuning/page.tsx, line 50:
<comment>The new Thresholds paragraph claims "An explicit 0 disables that gate entirely," but that is false for scenario_dedupe. The backend clamps a non-positive dedupe threshold to the 0.95 default (backend/internal/pipeline/scenarios.go scenarioDedupeThreshold), because similarity >= 0 is always true and a 0 would otherwise suppress every new scenario — the opposite of disabling the gate. Qualify the statement so a user setting scenario_dedupe to 0 is not misled into expecting dedup off.</comment>
<file context>
@@ -52,92 +44,86 @@ export default function MemoryTuningPage() {
+ more relevant matches). Lower = more permissive (more context, more noise). Scores are raw
+ cosine similarity over the configured embedding space — they sit high, so the defaults
+ are higher than intuition suggests. An explicit{" "}
+ <code className="bg-slate-900 px-1 text-amber">0</code> disables that gate entirely.
</p>
</file context>
| <code className="bg-slate-900 px-1 text-amber">0</code> disables that gate entirely. | |
| <code className="bg-slate-900 px-1 text-amber">0</code> disables the similarity gates — except <code className="bg-slate-900 px-1 text-amber">scenario_dedupe</code>, where 0 is clamped back to the default 0.95 because a zero floor would otherwise treat every candidate as a duplicate. |
| Ticking the box requires <strong className="text-slate-200">repository | ||
| write access</strong> — Argus checks the editor's permission, not | ||
| just the comment body. The review runs under a tighter{" "} | ||
| <strong className="text-slate-200">10/hour per-repo</strong> cap for |
There was a problem hiding this comment.
P2: The FAQ now states checkbox-triggered reviews run under a 10/hour per-repo force bucket, but backend/internal/api/handlers_webhook.go:424-428 still documents the same dispatch flow as "the tighter force=true rate-limit path (3/hr/repo)". Verify the actual admission bucket size before rollout: if the limit is still 3/hr, these docs (and the matching claims in docs-content.tsx) repeat the misstatement this PR set out to fix; if it is 10/hr, the backend comment is stale and should be corrected alongside.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/src/app/docs/faq/page.tsx, line 68:
<comment>The FAQ now states checkbox-triggered reviews run under a 10/hour per-repo force bucket, but backend/internal/api/handlers_webhook.go:424-428 still documents the same dispatch flow as "the tighter `force=true` rate-limit path (3/hr/repo)". Verify the actual admission bucket size before rollout: if the limit is still 3/hr, these docs (and the matching claims in docs-content.tsx) repeat the misstatement this PR set out to fix; if it is 10/hr, the backend comment is stale and should be corrected alongside.</comment>
<file context>
@@ -62,10 +62,11 @@ export default function FAQPage() {
+ Ticking the box requires <strong className="text-slate-200">repository
+ write access</strong> — Argus checks the editor's permission, not
+ just the comment body. The review runs under a tighter{" "}
+ <strong className="text-slate-200">10/hour per-repo</strong> cap for
+ checkbox-triggered reviews (same bucket as the{" "}
<code className="bg-slate-900 px-1 text-amber">--force</code> command flag).
</file context>
| "On by default. When off, Argus posts a task-list checkbox on opened PRs (and on new pushes) with an estimated token / cost preview. Ticking the box runs the review on demand.", | ||
| defaultValue: true, | ||
| "Off by default on hosted Argus (self-hosted defaults to on). When off, Argus posts a task-list checkbox on opened PRs (and on new pushes) with an estimated token / cost preview. Ticking the box runs the review on demand.", | ||
| defaultValue: false, |
There was a problem hiding this comment.
P3: On self-hosted installs the Auto-review toggle now renders OFF by default even though reviews still run unconditionally. The new description says "self-hosted defaults to on", but the toggle's displayed value comes only from the stored setting or AUTO_RUN_TOGGLE.defaultValue (line 1465-1467 / 1915-1919), and the single defaultValue=false cannot reflect deployment type. Backend decideAutoRun/AutoRun short-circuits self-hosted to ALWAYS review regardless of stored settings (backend/internal/pipeline/persona.go), so a fresh self-hosted org shows "Auto-review every PR" off in the settings page while the backend keeps auto-running reviews — contradicting the new description. Previously defaultValue=true at least displayed self-hosted's actual behavior. Consider deriving the shown default from a self-hosted flag when the backend exposes one, or narrow the description wording.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/src/app/(dashboard)/settings/page.tsx, line 922:
<comment>On self-hosted installs the Auto-review toggle now renders OFF by default even though reviews still run unconditionally. The new description says "self-hosted defaults to on", but the toggle's displayed value comes only from the stored setting or AUTO_RUN_TOGGLE.defaultValue (line 1465-1467 / 1915-1919), and the single defaultValue=false cannot reflect deployment type. Backend decideAutoRun/AutoRun short-circuits self-hosted to ALWAYS review regardless of stored settings (backend/internal/pipeline/persona.go), so a fresh self-hosted org shows "Auto-review every PR" off in the settings page while the backend keeps auto-running reviews — contradicting the new description. Previously defaultValue=true at least displayed self-hosted's actual behavior. Consider deriving the shown default from a self-hosted flag when the backend exposes one, or narrow the description wording.</comment>
<file context>
@@ -918,8 +918,8 @@ const AUTO_RUN_TOGGLE = {
- "On by default. When off, Argus posts a task-list checkbox on opened PRs (and on new pushes) with an estimated token / cost preview. Ticking the box runs the review on demand.",
- defaultValue: true,
+ "Off by default on hosted Argus (self-hosted defaults to on). When off, Argus posts a task-list checkbox on opened PRs (and on new pushes) with an estimated token / cost preview. Ticking the box runs the review on demand.",
+ defaultValue: false,
} as const;
</file context>
| | Scenario memory | ON | Surfaces past bugs when you touch the same files | | ||
| | Code simulation | OFF | Tests known failure scenarios against your diff | | ||
| | Auto-review every PR | Off (hosted) / on (self-hosted unset) | Reviews run automatically vs. on-demand via checkbox | | ||
| | Auto-resolve stale threads | On | Pushes that fix flagged lines resolve the thread — no LLM call | |
There was a problem hiding this comment.
P3: The claim that auto-resolve involves "no LLM call" contradicts the codebase: the auto-resolve path runs an addressed-judge LLM pass that consumes model tokens (backend/internal/pipeline/auto_resolve_tokens_test.go records judge prompt/model/cost tokens), and README/CONTEXT both state "stale threads auto-resolve only after LLM judge confirms the change addressed the finding; line proximity alone never resolves" and "Proximity is prefilter only; addressed-judge must confirm fix." So a push that merely changes the anchored lines does not itself close the thread — a judge call is involved and the thread stays open on judge failure. Please reconcile the wording (drop "no LLM call" and the implication that the push alone resolves) with the actual autoResolveOnSynchronize behavior.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/public/docs.md, line 245:
<comment>The claim that auto-resolve involves "no LLM call" contradicts the codebase: the auto-resolve path runs an addressed-judge LLM pass that consumes model tokens (`backend/internal/pipeline/auto_resolve_tokens_test.go` records judge prompt/model/cost tokens), and README/CONTEXT both state "stale threads auto-resolve only after LLM judge confirms the change addressed the finding; line proximity alone never resolves" and "Proximity is prefilter only; addressed-judge must confirm fix." So a push that merely changes the anchored lines does not itself close the thread — a judge call is involved and the thread stays open on judge failure. Please reconcile the wording (drop "no LLM call" and the implication that the push alone resolves) with the actual `autoResolveOnSynchronize` behavior.</comment>
<file context>
@@ -178,55 +204,74 @@ Example: *"All API endpoints must validate authentication tokens before processi
-| Scenario memory | ON | Surfaces past bugs when you touch the same files |
-| Code simulation | OFF | Tests known failure scenarios against your diff |
+| Auto-review every PR | Off (hosted) / on (self-hosted unset) | Reviews run automatically vs. on-demand via checkbox |
+| Auto-resolve stale threads | On | Pushes that fix flagged lines resolve the thread — no LLM call |
+| Deep Review | Off | 4 specialist reviewers per file; required for simulation |
+| Simulation & Scenarios | Off | Tests stored failure scenarios against the diff |
</file context>
| testGeneration: true, | ||
| ideIntegration: false, | ||
| ticketing: false, | ||
| ticketing: true, |
There was a problem hiding this comment.
P3: Greptile's ticketing flag was flipped to true while the same entry's strengths and summary say ticket handling is context-only: "Reviews code against linked Jira tickets, Linear issues, and Confluence pages (read-only — no ticket creation)" and "doesn't ... create tickets." Every other entry sets ticketing: true only where ticket creation exists (Cubic "creates Jira/Linear tickets", Codacy "creates tickets from findings", SonarQube/Gitar "creates follow-up issues"), while Greptile explicitly does not create tickets. If the flag means ticket creation/integration uniformly, this contradicts the copy; if it now means context-only, its semantics differ from the other entries. Align the flag with the written behavior or the other entries.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At web/src/lib/pseo/competitors.ts, line 174:
<comment>Greptile's `ticketing` flag was flipped to true while the same entry's strengths and summary say ticket handling is context-only: "Reviews code against linked Jira tickets, Linear issues, and Confluence pages (read-only — no ticket creation)" and "doesn't ... create tickets." Every other entry sets `ticketing: true` only where ticket creation exists (Cubic "creates Jira/Linear tickets", Codacy "creates tickets from findings", SonarQube/Gitar "creates follow-up issues"), while Greptile explicitly does not create tickets. If the flag means ticket creation/integration uniformly, this contradicts the copy; if it now means context-only, its semantics differ from the other entries. Align the flag with the written behavior or the other entries.</comment>
<file context>
@@ -168,87 +171,89 @@ export const competitors: Competitor[] = [
testGeneration: true,
ideIntegration: false,
- ticketing: false,
+ ticketing: true,
complianceCert: true,
},
</file context>
| ticketing: true, | |
| ticketing: false, |
Summary
MEMORY_DEFAULTSmirrored stale backend thresholds;ratelimit.gocomments listed wrong limits;auto_runtoggle claimed "on by default" while hosted treats unset as off.Test plan
pnpm lint— no new warnings in touched filespnpm typecheck— cleanpnpm build— all 69 pages, all 10/compare/*routes (verified on private-base tree)Summary by cubic
Refreshes competitor comparisons and docs to match current pricing, features, and actual codebase behavior.
Docs
/compareentries against official pricing pages; updates plan renames, credit pricing, and new capabilities.Bug fixes
MEMORY_DEFAULTSnow matches backend thresholds.ratelimit.gocomments now list correct limits.Written for commit dd7d36f. Summary will update on new commits.