co-maintainer 0.7.2-beta.1 → 0.7.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/package.json +2 -1
- package/dist/src/ai/anthropic.d.ts +1 -1
- package/dist/src/ai/anthropic.js +1 -1
- package/dist/src/ai/batch.d.ts +1 -1
- package/dist/src/ai/batch.js +3 -3
- package/dist/src/ai/copy_rule.d.ts +7 -0
- package/dist/src/ai/copy_rule.js +23 -0
- package/dist/src/ai/estimate.js +3 -3
- package/dist/src/ai/fake.js +2 -1
- package/dist/src/ai/mermaid.js +1 -0
- package/dist/src/ai/oauth/registry.d.ts +1 -1
- package/dist/src/ai/oauth/sign-in-with-chatgpt.d.ts +4 -4
- package/dist/src/ai/oauth/sign-in-with-chatgpt.js +6 -6
- package/dist/src/ai/openai.d.ts +3 -3
- package/dist/src/ai/openai.js +4 -4
- package/dist/src/ai/openrouter.d.ts +1 -1
- package/dist/src/ai/openrouter.js +1 -1
- package/dist/src/ai/pricing.js +1 -1
- package/dist/src/ai/provider.d.ts +2 -0
- package/dist/src/ai/provider.js +8 -2
- package/dist/src/ai/verify.d.ts +1 -1
- package/dist/src/ai/verify.js +1 -1
- package/dist/src/cli/args.js +1 -1
- package/dist/src/cli/command_json.d.ts +4 -2
- package/dist/src/cli/command_json.js +1 -0
- package/dist/src/cli/commands/config.d.ts +1 -1
- package/dist/src/cli/commands/config.js +1 -1
- package/dist/src/cli/commands/oauth.js +1 -1
- package/dist/src/cli/commands/probe.js +1 -1
- package/dist/src/cli/commands/review.js +2 -0
- package/dist/src/cli/commands/serve.d.ts +1 -1
- package/dist/src/cli/commands/serve.js +1 -1
- package/dist/src/cli/commands/set.d.ts +1 -1
- package/dist/src/cli/commands/set.js +23 -6
- package/dist/src/cli/commands/view.js +2 -2
- package/dist/src/cli/error.d.ts +1 -1
- package/dist/src/cli/error.js +1 -1
- package/dist/src/cli/pr_carry.d.ts +2 -0
- package/dist/src/cli/pr_carry.js +2 -2
- package/dist/src/cli/prompt.d.ts +2 -2
- package/dist/src/cli/prompt.js +3 -3
- package/dist/src/cli/review_args.js +2 -2
- package/dist/src/cli/review_carry.d.ts +3 -1
- package/dist/src/cli/review_carry.js +5 -3
- package/dist/src/cli/review_result.d.ts +13 -5
- package/dist/src/cli/review_result.js +19 -9
- package/dist/src/config.d.ts +20 -11
- package/dist/src/config.js +3 -3
- package/dist/src/github/app.d.ts +1 -1
- package/dist/src/github/app.js +1 -1
- package/dist/src/github/app_manifest.d.ts +1 -1
- package/dist/src/github/app_manifest.js +1 -1
- package/dist/src/github/client.d.ts +1 -1
- package/dist/src/github/client.js +1 -1
- package/dist/src/github/collect.d.ts +7 -1
- package/dist/src/github/collect.js +41 -16
- package/dist/src/github/gh.js +2 -2
- package/dist/src/github/jwt.d.ts +1 -1
- package/dist/src/github/jwt.js +1 -1
- package/dist/src/github/oauth.js +2 -2
- package/dist/src/knowledge/facts.js +3 -3
- package/dist/src/knowledge/guide.d.ts +24 -1
- package/dist/src/knowledge/guide.js +46 -16
- package/dist/src/knowledge/guide_ids.d.ts +9 -0
- package/dist/src/knowledge/guide_ids.js +26 -7
- package/dist/src/knowledge/probe.js +1 -1
- package/dist/src/knowledge/rule_changes.d.ts +18 -0
- package/dist/src/knowledge/rule_changes.js +41 -0
- package/dist/src/knowledge/rules.d.ts +11 -0
- package/dist/src/knowledge/rules.js +10 -1
- package/dist/src/knowledge/skill.js +1 -1
- package/dist/src/knowledge/stable_ids.d.ts +29 -0
- package/dist/src/knowledge/stable_ids.js +90 -0
- package/dist/src/knowledge/synthesis.d.ts +13 -1
- package/dist/src/knowledge/synthesis.js +94 -26
- package/dist/src/knowledge/types.d.ts +16 -0
- package/dist/src/local/carry_over_store.d.ts +1 -1
- package/dist/src/local/carry_over_store.js +1 -1
- package/dist/src/local/codegraph_prepare.js +1 -1
- package/dist/src/local/git_ops.js +1 -1
- package/dist/src/local/review_local.js +6 -5
- package/dist/src/local/review_lock.d.ts +1 -1
- package/dist/src/pr/checkout.d.ts +3 -3
- package/dist/src/pr/checkout.js +7 -7
- package/dist/src/pr/codegraph_tool_args.d.ts +1 -1
- package/dist/src/pr/codegraph_tool_args.js +1 -1
- package/dist/src/pr/findings.d.ts +3 -0
- package/dist/src/pr/findings.js +29 -10
- package/dist/src/pr/findings_json.d.ts +2 -2
- package/dist/src/pr/findings_json.js +15 -2
- package/dist/src/pr/github_revision.js +1 -1
- package/dist/src/pr/review_copy.d.ts +2 -3
- package/dist/src/pr/review_copy.js +2 -3
- package/dist/src/pr/reviewer.d.ts +7 -3
- package/dist/src/pr/reviewer.js +26 -26
- package/dist/src/pr/scope.d.ts +8 -8
- package/dist/src/pr/scope.js +4 -4
- package/dist/src/pr/snapshot.d.ts +2 -2
- package/dist/src/remote/client.js +4 -3
- package/dist/src/remote/schema.d.ts +1 -1
- package/dist/src/remote/schema.js +1 -1
- package/dist/src/remote/server/body.d.ts +1 -1
- package/dist/src/remote/server/body.js +1 -1
- package/dist/src/remote/server/guides.js +1 -1
- package/dist/src/remote/validate.js +1 -1
- package/dist/src/review/blocking.d.ts +3 -3
- package/dist/src/review/blocking.js +2 -2
- package/dist/src/review/carry_over.d.ts +9 -1
- package/dist/src/review/carry_over.js +10 -2
- package/dist/src/review/diagram.js +7 -6
- package/dist/src/review/engine.d.ts +1 -1
- package/dist/src/review/guides.js +1 -1
- package/dist/src/review/references.d.ts +68 -0
- package/dist/src/review/references.js +176 -0
- package/dist/src/review/system_prompt.d.ts +1 -1
- package/dist/src/review/system_prompt.js +1 -1
- package/dist/src/server/api/jobs.js +2 -2
- package/dist/src/server/api/repos.js +1 -1
- package/dist/src/server/app.js +1 -1
- package/dist/src/server/auth.d.ts +3 -3
- package/dist/src/server/auth.js +9 -4
- package/dist/src/server/http.js +5 -5
- package/dist/src/server/pages/activity.js +1 -1
- package/dist/src/server/pages/add_repo.js +1 -0
- package/dist/src/server/pages/client.d.ts +2 -2
- package/dist/src/server/pages/client.js +2 -2
- package/dist/src/server/pages/home.d.ts +1 -1
- package/dist/src/server/pages/home.js +1 -1
- package/dist/src/server/pages/icons.d.ts +1 -1
- package/dist/src/server/pages/icons.js +1 -1
- package/dist/src/server/pages/layout.d.ts +2 -2
- package/dist/src/server/pages/layout.js +3 -3
- package/dist/src/server/pages/pr.js +4 -0
- package/dist/src/server/pages/rate_limit_waits.js +2 -2
- package/dist/src/server/pages/references.d.ts +2 -0
- package/dist/src/server/pages/references.js +55 -0
- package/dist/src/server/pages/repo.js +26 -0
- package/dist/src/server/pages/repo_settings.js +1 -0
- package/dist/src/server/pages/router.js +3 -2
- package/dist/src/server/pages/styles.d.ts +2 -2
- package/dist/src/server/pages/styles.js +2 -2
- package/dist/src/server/pages/ui.d.ts +1 -1
- package/dist/src/server/pages/ui.js +2 -2
- package/dist/src/server/webhook/index.js +1 -1
- package/dist/src/services/clear.d.ts +2 -2
- package/dist/src/services/clear.js +2 -2
- package/dist/src/services/dashboard.d.ts +2 -0
- package/dist/src/services/dashboard.js +2 -0
- package/dist/src/services/probe.js +1 -1
- package/dist/src/services/push_review.js +4 -1
- package/dist/src/services/remote_review.js +5 -2
- package/dist/src/services/remote_review_abort.d.ts +1 -1
- package/dist/src/services/remote_review_abort.js +1 -1
- package/dist/src/services/remote_tokens.js +1 -1
- package/dist/src/services/replies.js +0 -2
- package/dist/src/services/review.d.ts +3 -5
- package/dist/src/services/review.js +43 -18
- package/dist/src/services/rules_build.d.ts +20 -1
- package/dist/src/services/rules_build.js +85 -6
- package/dist/src/services/setup.d.ts +2 -2
- package/dist/src/services/setup.js +36 -11
- package/dist/src/services/watch_comments.js +1 -2
- package/dist/src/services/watch_notes.js +1 -3
- package/dist/src/store/app_db.d.ts +1 -1
- package/dist/src/store/app_db.js +2 -2
- package/dist/src/store/findings.d.ts +2 -0
- package/dist/src/store/findings.js +3 -3
- package/dist/src/store/jobs.d.ts +2 -2
- package/dist/src/store/jobs.js +3 -3
- package/dist/src/store/migrations.d.ts +2 -2
- package/dist/src/store/migrations.js +35 -18
- package/dist/src/store/rows.d.ts +4 -2
- package/dist/src/store/rows.js +1 -1
- package/dist/src/store/rule_events.d.ts +15 -0
- package/dist/src/store/rule_events.js +20 -0
- package/dist/src/store/sessions.d.ts +1 -1
- package/dist/src/store/sessions.js +2 -2
- package/dist/src/store/sqlite.js +2 -2
- package/dist/src/tools/codegraph.d.ts +1 -1
- package/dist/src/tools/codegraph.js +3 -3
- package/dist/src/tools/codegraph_exec.d.ts +1 -1
- package/dist/src/tools/codegraph_exec.js +1 -1
- package/dist/src/types.d.ts +6 -2
- package/dist/src/util/keyed_lock.d.ts +1 -1
- package/dist/src/util/keyed_lock.js +1 -1
- package/dist/src/util/log.js +1 -1
- package/dist/src/util/redact.js +2 -1
- package/dist/src/util/run_summary.d.ts +3 -3
- package/dist/src/util/run_summary.js +3 -3
- package/dist/src/util/runtime.d.ts +4 -4
- package/dist/src/util/runtime.js +5 -5
- package/dist/src/util/time.d.ts +1 -1
- package/dist/src/util/time.js +1 -1
- package/package.json +2 -1
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
import { type ToolHandler } from "../ai/mermaid_loop.ts";
|
|
2
|
+
import type { ReferenceContext } from "../review/references.ts";
|
|
2
3
|
import type { Revision } from "../review/revision.ts";
|
|
3
4
|
import { type Coverage, type CoverageWarning } from "../review/coverage.ts";
|
|
4
5
|
import type { Snapshot } from "./snapshot.ts";
|
|
@@ -40,6 +41,9 @@ export type ReviewOutcome = AiResponse & {
|
|
|
40
41
|
* the files it skipped as unchanged. */
|
|
41
42
|
visiblePaths: string[];
|
|
42
43
|
guideBuiltAt: string | null;
|
|
44
|
+
/** What the finding references resolve against: the `RULES.json` and the
|
|
45
|
+
* codebase conventions this review was given. Never sent to the model. */
|
|
46
|
+
referenceContext: ReferenceContext;
|
|
43
47
|
codegraphState: "used" | "disabled" | "unavailable";
|
|
44
48
|
coverage: Coverage;
|
|
45
49
|
/** `diff_dropped`, `diff_withheld` and `summary_failed`. */
|
|
@@ -55,7 +59,7 @@ export type ReviewOutcome = AiResponse & {
|
|
|
55
59
|
* the review itself. A server with no low model configured summarized with
|
|
56
60
|
* the review's provider before, and still does, so no review fails for it. */
|
|
57
61
|
export declare function summaryProvider(options: Options, reviewAi: AiProvider, report: ProgressSink, extras?: ReviewExtras): AiProvider;
|
|
58
|
-
export declare const MERMAID_GUIDANCE = "Mermaid selection and minimal syntax:\nflowchart = decisions, branches, pipelines, and fallback paths
|
|
62
|
+
export declare const MERMAID_GUIDANCE = "Mermaid selection and minimal syntax:\nflowchart = decisions, branches, pipelines, and fallback paths.\nswimlane-beta = work crossing owners, actors, services, teams, or layers.\nsequenceDiagram = ordered calls, webhooks, retries, responses, and timing.\nclassDiagram = class, interface, type, inheritance, or composition relationships.\nstateDiagram-v2 = lifecycle and state transitions.\nerDiagram = database entities, keys, and cardinality.\nrequirementDiagram = requirements linked to tests or implementation.\nusecase-beta = actors and system capabilities.\nC4Context = users, systems, boundaries, and integrations.\nzenuml = compact nested call sequences.\npacket = binary fields, bit ranges, and protocol layout.\narchitecture-beta = services, containers, storage, and deployment topology.\neventmodeling = commands, events, processors, read models, and timelines.\ntreeView-beta = directory, file, module, or dependency hierarchy.\nIf a diagram is useful, call read-mermaid-syntaxes before writing any Mermaid.\nRequest every type you need in a single call, because repeated tool rounds are\ncapped. Use the smallest type that matches the evidence.";
|
|
59
63
|
/** Without tool support the model cannot read the Mermaid syntax docs, so
|
|
60
64
|
* asking for a diagram only invites invented syntax. */
|
|
61
65
|
export declare const NO_DIAGRAM_RULES = "Do not use Mermaid or any other diagram. Explain with prose only.";
|
|
@@ -63,7 +67,7 @@ export declare function reviewSystemPrompt(diagrams: boolean, options?: Options)
|
|
|
63
67
|
/** The guide block every review sends, kept as the stable head of the system
|
|
64
68
|
* prompt rather than the user prompt so a provider with prompt caching
|
|
65
69
|
* (Anthropic's `cache_control`) reads it instead of re-billing it. Nothing
|
|
66
|
-
* time-varying may enter this string
|
|
70
|
+
* time-varying may enter this string, a single differing byte makes the cache
|
|
67
71
|
* miss. */
|
|
68
72
|
export declare function reviewGuidesBlock(guide: string, detailed: string, codebase: string, instructions?: string): string;
|
|
69
73
|
/** Shared PR and local/remote workspace review instructions: confirm claims
|
|
@@ -74,5 +78,5 @@ export declare function clampImproveMatrix(matrix: number): number;
|
|
|
74
78
|
export declare function readGuide(repo: string, name: string): Promise<string>;
|
|
75
79
|
export type { Snapshot };
|
|
76
80
|
export declare function reviewPullRequest(client: GitHubClient, options: Options, usage?: UsageSink, snapshot?: Snapshot, ai?: AiProvider, progress?: ProgressSink, extras?: ReviewExtras): Promise<ReviewOutcome>;
|
|
77
|
-
/** Local workspace review
|
|
81
|
+
/** Local workspace review, same prompt loop as PR review without GitHub. */
|
|
78
82
|
export declare function reviewWorkspaceRevision(revision: Revision, options: Options, headSha: string, usage?: UsageSink, ai?: AiProvider, progress?: ProgressSink, extras?: ReviewExtras): Promise<ReviewOutcome>;
|
package/dist/src/pr/reviewer.js
CHANGED
|
@@ -5,7 +5,7 @@ import { loadGuides } from "../review/guides.js";
|
|
|
5
5
|
import { notInitializedError } from "../review/not_initialized.js";
|
|
6
6
|
import { computeScope } from "./scope.js";
|
|
7
7
|
import { prepareCodegraphTools } from "./codegraph_tools.js";
|
|
8
|
-
import {
|
|
8
|
+
import { severitySection } from "./review_copy.js";
|
|
9
9
|
import { repositoryInstructionsBlock, reviewSystemPromptBlock, } from "../review/system_prompt.js";
|
|
10
10
|
import { readDiffMaxChars } from "./diff_summary.js";
|
|
11
11
|
import { intentField, intentPromptNote } from "../review/intent.js";
|
|
@@ -78,19 +78,19 @@ function diffReport(chars, coverage) {
|
|
|
78
78
|
return `diff prepared · ${chars} chars${coverageTruncated(coverage) ? ` · ${coverageCounts(coverage)}` : ""}`;
|
|
79
79
|
}
|
|
80
80
|
export const MERMAID_GUIDANCE = `Mermaid selection and minimal syntax:
|
|
81
|
-
flowchart = decisions, branches, pipelines, and fallback paths
|
|
82
|
-
swimlane-beta = work crossing owners, actors, services, teams, or layers
|
|
83
|
-
sequenceDiagram = ordered calls, webhooks, retries, responses, and timing
|
|
84
|
-
classDiagram = class, interface, type, inheritance, or composition relationships
|
|
85
|
-
stateDiagram-v2 = lifecycle and state transitions
|
|
86
|
-
erDiagram = database entities, keys, and cardinality
|
|
87
|
-
requirementDiagram = requirements linked to tests or implementation
|
|
88
|
-
usecase-beta = actors and system capabilities
|
|
89
|
-
C4Context = users, systems, boundaries, and integrations
|
|
90
|
-
zenuml = compact nested call sequences
|
|
91
|
-
packet = binary fields, bit ranges, and protocol layout
|
|
92
|
-
architecture-beta = services, containers, storage, and deployment topology
|
|
93
|
-
eventmodeling = commands, events, processors, read models, and timelines
|
|
81
|
+
flowchart = decisions, branches, pipelines, and fallback paths.
|
|
82
|
+
swimlane-beta = work crossing owners, actors, services, teams, or layers.
|
|
83
|
+
sequenceDiagram = ordered calls, webhooks, retries, responses, and timing.
|
|
84
|
+
classDiagram = class, interface, type, inheritance, or composition relationships.
|
|
85
|
+
stateDiagram-v2 = lifecycle and state transitions.
|
|
86
|
+
erDiagram = database entities, keys, and cardinality.
|
|
87
|
+
requirementDiagram = requirements linked to tests or implementation.
|
|
88
|
+
usecase-beta = actors and system capabilities.
|
|
89
|
+
C4Context = users, systems, boundaries, and integrations.
|
|
90
|
+
zenuml = compact nested call sequences.
|
|
91
|
+
packet = binary fields, bit ranges, and protocol layout.
|
|
92
|
+
architecture-beta = services, containers, storage, and deployment topology.
|
|
93
|
+
eventmodeling = commands, events, processors, read models, and timelines.
|
|
94
94
|
treeView-beta = directory, file, module, or dependency hierarchy.
|
|
95
95
|
If a diagram is useful, call read-mermaid-syntaxes before writing any Mermaid.
|
|
96
96
|
Request every type you need in a single call, because repeated tool rounds are
|
|
@@ -131,7 +131,7 @@ ${diagrams ? (plantumlSystemRules(options) ?? REVIEW_DIAGRAM_RULES) : NO_DIAGRAM
|
|
|
131
131
|
/** The guide block every review sends, kept as the stable head of the system
|
|
132
132
|
* prompt rather than the user prompt so a provider with prompt caching
|
|
133
133
|
* (Anthropic's `cache_control`) reads it instead of re-billing it. Nothing
|
|
134
|
-
* time-varying may enter this string
|
|
134
|
+
* time-varying may enter this string, a single differing byte makes the cache
|
|
135
135
|
* miss. */
|
|
136
136
|
export function reviewGuidesBlock(guide, detailed, codebase, instructions = "") {
|
|
137
137
|
return `REVIEW GUIDE:
|
|
@@ -158,7 +158,7 @@ when a tool can answer. A missing regression test is not a substitute for
|
|
|
158
158
|
identifying the concrete input or code path that misbehaves when you can.`;
|
|
159
159
|
/** Turns a model reply into the Markdown every consumer already parses.
|
|
160
160
|
* A reply that is not usable JSON is retried once with an
|
|
161
|
-
* explicit reminder
|
|
161
|
+
* explicit reminder, if that also fails, the raw text is returned so the
|
|
162
162
|
* legacy Markdown parser can still read it. This keeps a provider that ignores
|
|
163
163
|
* `response_format` working, and never drops a review on the floor. */
|
|
164
164
|
async function normalizeReviewResponse(provider, request, response, extraTools, maxToolRounds, usage, report, options) {
|
|
@@ -255,9 +255,9 @@ export async function reviewPullRequest(client, options, usage, snapshot, ai, pr
|
|
|
255
255
|
console.log(`[debug] PR #${number} · comments=${comments.length} · reviews=${reviews.length} · files=${files.length}`);
|
|
256
256
|
}
|
|
257
257
|
// What the author actually wrote this round, as opposed to code that
|
|
258
|
-
// arrived by merging the default branch in
|
|
258
|
+
// arrived by merging the default branch in, see scope.ts. A re-review
|
|
259
259
|
// round's diff regularly carries a `merge main` that dwarfs the PR's own
|
|
260
|
-
// change and that no human reviewer reads either
|
|
260
|
+
// change and that no human reviewer reads either, --review-upstream turns
|
|
261
261
|
// this off and reviews everything, matching the pre-scope behavior.
|
|
262
262
|
const headSha = snapshot?.commit ?? String(pr.head?.sha ?? "");
|
|
263
263
|
const baseRevision = snapshot?.base ?? String(pr.base?.ref ?? "main");
|
|
@@ -309,7 +309,7 @@ export async function reviewPullRequest(client, options, usage, snapshot, ai, pr
|
|
|
309
309
|
report(`codegraph tools · ${codegraphTools.length > 0 ? "ready" : "unavailable"}`);
|
|
310
310
|
}
|
|
311
311
|
const extraTools = [...codegraphTools, ...prepared.tools];
|
|
312
|
-
// Bigger own-scope diffs need more codegraph round-trips to trace
|
|
312
|
+
// Bigger own-scope diffs need more codegraph round-trips to trace, fixed
|
|
313
313
|
// at 3 rounds regardless of size caused the model to run out mid-review on
|
|
314
314
|
// large PRs (verified via a tokio benchmark run: it hallucinated fake
|
|
315
315
|
// tool-call text instead of a finding once tools were dropped).
|
|
@@ -346,12 +346,12 @@ Do not invent requirements. Ignore bot noise and historical PR identities.
|
|
|
346
346
|
Reason thoroughly, then answer with the JSON object described below and nothing
|
|
347
347
|
else. There is no fixed number of findings. Return every independently
|
|
348
348
|
actionable finding supported by the diff and guide, including zero findings
|
|
349
|
-
when appropriate. Do not stop early
|
|
349
|
+
when appropriate. Do not stop early. Inspect all supplied diff text first and
|
|
350
350
|
return the natural count. If the diff contains a truncation marker, limit
|
|
351
351
|
claims to the supplied text and do not imply that omitted files were reviewed.
|
|
352
352
|
Do not invent low-value findings.
|
|
353
353
|
When the DIFF section below has an UPSTREAM CONTEXT part, that code arrived
|
|
354
|
-
through a merge and was not authored by this pull request
|
|
354
|
+
through a merge and was not authored by this pull request. Do not raise a
|
|
355
355
|
finding located only there, and never mark blocking a finding whose only
|
|
356
356
|
support is upstream context.
|
|
357
357
|
${CODEGRAPH_DIFF_VERIFICATION}
|
|
@@ -360,7 +360,6 @@ suggestion. Write prose only: do not emit Markdown headings, a Location line,
|
|
|
360
360
|
backticks around the path, or the sentence "If you'd like me to explain it in
|
|
361
361
|
more detail, please ask." Our code renders the heading, the location, and the
|
|
362
362
|
suggestion from your JSON fields. Use P0-P3 severity, and true for blocking.
|
|
363
|
-
${REVIEW_COPY_RULE}
|
|
364
363
|
Every diff line in the DIFF section starts with its line number in the new
|
|
365
364
|
file. Copy "lineFrom" and "lineTo" from that column instead of counting from
|
|
366
365
|
the @@ header. Removed lines have no number, so anchor a finding about removed
|
|
@@ -409,7 +408,7 @@ ${diff}`;
|
|
|
409
408
|
if (!response.text.trim()) {
|
|
410
409
|
throw new Error(`${aiProviderName(options)} returned an empty review. The reasoning budget may have been exhausted`);
|
|
411
410
|
}
|
|
412
|
-
// The model returns JSON
|
|
411
|
+
// The model returns JSON. We render the Markdown. A reply
|
|
413
412
|
// that is not usable JSON is retried once and then falls back to the legacy
|
|
414
413
|
// Markdown parser, so a provider that ignores the schema still works.
|
|
415
414
|
let reviewText = await normalizeReviewResponse(provider, request, response, extraTools, maxToolRounds, usage, report, options);
|
|
@@ -452,6 +451,7 @@ ${reviewText}`,
|
|
|
452
451
|
text: `${severitySection()}\n${renderDiagrams(reviewText, options)}`,
|
|
453
452
|
visiblePaths: prepared.seenPaths,
|
|
454
453
|
guideBuiltAt: guides.guideBuiltAt,
|
|
454
|
+
referenceContext: { rules: guides.rules, codebase: guides.codebase },
|
|
455
455
|
codegraphState: !options.useCodegraph
|
|
456
456
|
? "disabled"
|
|
457
457
|
: codegraphTools.length > 0
|
|
@@ -462,7 +462,7 @@ ${reviewText}`,
|
|
|
462
462
|
diffStats: diffStats(named.map(({ file }) => diffFileFromGithub(file))),
|
|
463
463
|
};
|
|
464
464
|
}
|
|
465
|
-
/** Local workspace review
|
|
465
|
+
/** Local workspace review, same prompt loop as PR review without GitHub. */
|
|
466
466
|
export async function reviewWorkspaceRevision(revision, options, headSha, usage, ai, progress, extras) {
|
|
467
467
|
const report = progress ?? (() => { });
|
|
468
468
|
report(modelSummary(options));
|
|
@@ -511,7 +511,6 @@ Do not invent low-value findings. If the diff contains a truncation marker or a
|
|
|
511
511
|
file is summarized, limit claims to the supplied text and use read-full-diff or
|
|
512
512
|
codegraph tools before asserting behavior outside what was shown.
|
|
513
513
|
${CODEGRAPH_DIFF_VERIFICATION}
|
|
514
|
-
${REVIEW_COPY_RULE}
|
|
515
514
|
${diagrams ? (plantumlPromptRules(options) ?? DIAGRAM_PROMPT_RULES) : NO_DIAGRAM_RULES}
|
|
516
515
|
${FINDINGS_JSON_INSTRUCTIONS}
|
|
517
516
|
|
|
@@ -549,7 +548,7 @@ ${ownDiff}${unchangedListing}`;
|
|
|
549
548
|
prompt: `Audit the draft review below against the complete diff and guides.
|
|
550
549
|
Preserve valid findings, correct inaccurate ones, remove duplicate or
|
|
551
550
|
unsupported ones, and add every missing actionable finding. Re-check each
|
|
552
|
-
finding with codegraph when it depends on behavior outside the diff
|
|
551
|
+
finding with codegraph when it depends on behavior outside the diff. Remove
|
|
553
552
|
findings that only looked plausible from the diff slice.
|
|
554
553
|
Return the complete revised review as the same JSON object, and nothing else.
|
|
555
554
|
|
|
@@ -574,6 +573,7 @@ ${reviewText}`,
|
|
|
574
573
|
text: `${severitySection()}\n${renderDiagrams(reviewText, options)}`,
|
|
575
574
|
visiblePaths: prepared.seenPaths,
|
|
576
575
|
guideBuiltAt: guides.guideBuiltAt,
|
|
576
|
+
referenceContext: { rules: guides.rules, codebase: guides.codebase },
|
|
577
577
|
codegraphState,
|
|
578
578
|
coverage: prepared.coverage,
|
|
579
579
|
warnings: prepared.warnings,
|
package/dist/src/pr/scope.d.ts
CHANGED
|
@@ -2,18 +2,18 @@ import { type Run } from "./checkout.ts";
|
|
|
2
2
|
/** What a pull request's author actually wrote this round, as opposed to code
|
|
3
3
|
* that arrived by merging the default branch in. A re-review round's diff
|
|
4
4
|
* regularly includes a `merge main` a human reviewer never reads and a bot
|
|
5
|
-
* should not either
|
|
5
|
+
* should not either, one measured case in this repo's own benchmark had a
|
|
6
6
|
* round diff of 48 files where the author's own commits touched 5. Findings
|
|
7
|
-
* belong in the own set
|
|
7
|
+
* belong in the own set. Upstream code is context, shown but not policed,
|
|
8
8
|
* because the author's own change can still interact badly with it. */
|
|
9
9
|
export type ScopeResult = {
|
|
10
10
|
/** Files touched by a commit the author made this round: their own
|
|
11
11
|
* non-merge commits, plus the conflict-resolution hunks of any merge they
|
|
12
|
-
* made. A clean "merge main in" contributes nothing here
|
|
12
|
+
* made. A clean "merge main in" contributes nothing here, `--cc` only
|
|
13
13
|
* surfaces lines that differ from every parent, which is exactly the part
|
|
14
14
|
* a human resolved by hand. */
|
|
15
15
|
ownFiles: Set<string>;
|
|
16
|
-
/** Everything else touched between the round's boundary and its head
|
|
16
|
+
/** Everything else touched between the round's boundary and its head,
|
|
17
17
|
* code that arrived via a merge and was not authored this round. */
|
|
18
18
|
upstreamFiles: Set<string>;
|
|
19
19
|
defaultBranch: string;
|
|
@@ -21,13 +21,13 @@ export type ScopeResult = {
|
|
|
21
21
|
mergeCommits: number;
|
|
22
22
|
};
|
|
23
23
|
/** The scope algorithm against a clone that is already on disk. Split out from
|
|
24
|
-
* `computeScope` so a caller that already manages its own clone
|
|
25
|
-
* benchmark's OCR runner has its own, separate from co-maintainer's
|
|
24
|
+
* `computeScope` so a caller that already manages its own clone, the
|
|
25
|
+
* benchmark's OCR runner has its own, separate from co-maintainer's, can run
|
|
26
26
|
* the identical algorithm without a second clone of the same repository. */
|
|
27
27
|
export declare function scopeInClone(clone: string, base: string, head: string, run?: Run, remote?: string): Promise<ScopeResult | undefined>;
|
|
28
28
|
/** Clones (or reuses) `repo` under co-maintainer's own cache and runs the
|
|
29
|
-
* scope algorithm against it. Returns `undefined` on any failure
|
|
30
|
-
* an unresolvable commit, no default branch to compare against
|
|
29
|
+
* scope algorithm against it. Returns `undefined` on any failure, no clone,
|
|
30
|
+
* an unresolvable commit, no default branch to compare against, so the
|
|
31
31
|
* caller's fallback is always "review everything, unscoped", never a broken
|
|
32
32
|
* review. */
|
|
33
33
|
export declare function computeScope(repo: string, base: string, head: string, options?: {
|
package/dist/src/pr/scope.js
CHANGED
|
@@ -8,8 +8,8 @@ function lines(output) {
|
|
|
8
8
|
.filter(Boolean);
|
|
9
9
|
}
|
|
10
10
|
/** The scope algorithm against a clone that is already on disk. Split out from
|
|
11
|
-
* `computeScope` so a caller that already manages its own clone
|
|
12
|
-
* benchmark's OCR runner has its own, separate from co-maintainer's
|
|
11
|
+
* `computeScope` so a caller that already manages its own clone, the
|
|
12
|
+
* benchmark's OCR runner has its own, separate from co-maintainer's, can run
|
|
13
13
|
* the identical algorithm without a second clone of the same repository. */
|
|
14
14
|
export async function scopeInClone(clone, base, head, run = runCommand, remote = "origin") {
|
|
15
15
|
let baseSha;
|
|
@@ -64,8 +64,8 @@ export async function scopeInClone(clone, base, head, run = runCommand, remote =
|
|
|
64
64
|
};
|
|
65
65
|
}
|
|
66
66
|
/** Clones (or reuses) `repo` under co-maintainer's own cache and runs the
|
|
67
|
-
* scope algorithm against it. Returns `undefined` on any failure
|
|
68
|
-
* an unresolvable commit, no default branch to compare against
|
|
67
|
+
* scope algorithm against it. Returns `undefined` on any failure, no clone,
|
|
68
|
+
* an unresolvable commit, no default branch to compare against, so the
|
|
69
69
|
* caller's fallback is always "review everything, unscoped", never a broken
|
|
70
70
|
* review. */
|
|
71
71
|
export async function computeScope(repo, base, head, options = {}) {
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
/** Pins a review to a historical diff/discussion state instead of the PR's
|
|
2
|
-
* current (latest) state
|
|
2
|
+
* current (latest) state, used by the benchmarks to show the model exactly
|
|
3
3
|
* what a human reviewer saw, without leaking later comments as hints.
|
|
4
4
|
* `base` defaults to the PR's merge-base with its base branch (the whole PR
|
|
5
|
-
* so far)
|
|
5
|
+
* so far). Pass an earlier review round's commit as `base` to show only the
|
|
6
6
|
* incremental diff a re-reviewer would see since that round (P11). */
|
|
7
7
|
export type Snapshot = {
|
|
8
8
|
base?: string;
|
|
@@ -162,7 +162,7 @@ export async function runRemoteReview(cli) {
|
|
|
162
162
|
gitRoot: root,
|
|
163
163
|
enabled: !cli.disableCodegraph,
|
|
164
164
|
allowInstall: cli.allowToolInstall,
|
|
165
|
-
// `--json` must stay machine-readable
|
|
165
|
+
// `--json` must stay machine-readable, `canPrompt` adds TTY and CI.
|
|
166
166
|
interactive: !cli.json && canPrompt(),
|
|
167
167
|
});
|
|
168
168
|
const localTools = toolHandlerMap(codegraphPrep.tools);
|
|
@@ -188,7 +188,7 @@ export async function runRemoteReview(cli) {
|
|
|
188
188
|
files: built.revision.files,
|
|
189
189
|
},
|
|
190
190
|
capabilities: { tools: capabilityTools },
|
|
191
|
-
// Only present when this run turned BYOK on
|
|
191
|
+
// Only present when this run turned BYOK on, `byokKey` is
|
|
192
192
|
// known to be set here because of the earlier `byokEnabled` guard.
|
|
193
193
|
...(byokEnabled ? { byok: { key: byokKey } } : {}),
|
|
194
194
|
// The client's one-run instructions, when given. The server applies
|
|
@@ -219,7 +219,7 @@ export async function runRemoteReview(cli) {
|
|
|
219
219
|
let afterLogSeq = 0;
|
|
220
220
|
let toolResults = [];
|
|
221
221
|
let toolCallsRun = 0;
|
|
222
|
-
// A trustworthy server drives at most a few tool rounds
|
|
222
|
+
// A trustworthy server drives at most a few tool rounds. Cap the total so a
|
|
223
223
|
// malicious one cannot make the client read its tree without limit.
|
|
224
224
|
const MAX_REMOTE_TOOL_CALLS = 200;
|
|
225
225
|
const intervalMs = hs.sync.intervalSeconds * 1000;
|
|
@@ -322,6 +322,7 @@ export async function runRemoteReview(cli) {
|
|
|
322
322
|
coverage: coverageFromResult(payload.result),
|
|
323
323
|
extensions,
|
|
324
324
|
intent: intent !== null,
|
|
325
|
+
showReferences: readConfig().repos?.[repo]?.reviewReferences === true,
|
|
325
326
|
}) +
|
|
326
327
|
"\n");
|
|
327
328
|
if (run)
|
|
@@ -10,7 +10,7 @@ export declare const REMOTE_SYNC_INTERVAL_SECONDS = 3;
|
|
|
10
10
|
* repository's own cap is applied on the server. Guards the body size. */
|
|
11
11
|
export declare const REMOTE_MAX_SYSTEM_PROMPT_CHARS = 200000;
|
|
12
12
|
/** SHA-256 of sorted `fixtures/v<N>/*.json` (name + contents, `\r\n` normalized
|
|
13
|
-
* to `\n`)
|
|
13
|
+
* to `\n`). Bump with schema version. */
|
|
14
14
|
export declare const REMOTE_FIXTURE_HASH: Record<number, string>;
|
|
15
15
|
export declare const REMOTE_SYNC_STATUSES: readonly ["queued", "running", "done", "failed", "canceled"];
|
|
16
16
|
export type RemoteSyncStatus = (typeof REMOTE_SYNC_STATUSES)[number];
|
|
@@ -10,7 +10,7 @@ export const REMOTE_SYNC_INTERVAL_SECONDS = 3;
|
|
|
10
10
|
* repository's own cap is applied on the server. Guards the body size. */
|
|
11
11
|
export const REMOTE_MAX_SYSTEM_PROMPT_CHARS = 200_000;
|
|
12
12
|
/** SHA-256 of sorted `fixtures/v<N>/*.json` (name + contents, `\r\n` normalized
|
|
13
|
-
* to `\n`)
|
|
13
|
+
* to `\n`). Bump with schema version. */
|
|
14
14
|
export const REMOTE_FIXTURE_HASH = {
|
|
15
15
|
1: "71e4e5f8637eab4b53d47817a688812a517310d967abe26aa0d7f4e7d7d65092",
|
|
16
16
|
};
|
|
@@ -14,7 +14,7 @@ export type BoundedBodyResult = {
|
|
|
14
14
|
reason: "too_large" | "empty";
|
|
15
15
|
};
|
|
16
16
|
/** Reads the raw body, streaming, and refuses once more than `maxBytes` have
|
|
17
|
-
* arrived
|
|
17
|
+
* arrived, this holds with no Content-Length (chunked) too. The caller gets
|
|
18
18
|
* the exact bytes, which the webhook HMAC needs. */
|
|
19
19
|
export declare function readBoundedBytes(request: Request, maxBytes: number): Promise<BoundedBytesResult>;
|
|
20
20
|
export declare function readBoundedUtf8(request: Request, maxBytes: number): Promise<BoundedBodyResult>;
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
/** Bounded request body read (bytes, streaming). */
|
|
2
2
|
/** Reads the raw body, streaming, and refuses once more than `maxBytes` have
|
|
3
|
-
* arrived
|
|
3
|
+
* arrived, this holds with no Content-Length (chunked) too. The caller gets
|
|
4
4
|
* the exact bytes, which the webhook HMAC needs. */
|
|
5
5
|
export async function readBoundedBytes(request, maxBytes) {
|
|
6
6
|
const lengthHeader = request.headers.get("content-length");
|
|
@@ -243,7 +243,7 @@ export function validateSubmitRequest(body) {
|
|
|
243
243
|
if (typeof tool.name !== "string" || !tool.name) {
|
|
244
244
|
return bad(`capabilities.tools[${i}].name`, "required");
|
|
245
245
|
}
|
|
246
|
-
// Unknown tools are ignored at runtime
|
|
246
|
+
// Unknown tools are ignored at runtime. No error here.
|
|
247
247
|
if (REMOTE_KNOWN_TOOL_NAMES.has(tool.name) &&
|
|
248
248
|
tool.version !== undefined &&
|
|
249
249
|
typeof tool.version !== "string") {
|
|
@@ -3,12 +3,12 @@
|
|
|
3
3
|
* The same pull request was reviewed twice, the same P2 finding came back
|
|
4
4
|
* `non-blocking` the first time and `blocking` the second, and the exit code
|
|
5
5
|
* changed because the model's own label moved. `model` keeps that behavior
|
|
6
|
-
* (the model decides) because it is what 0.4.13 shipped
|
|
6
|
+
* (the model decides) because it is what 0.4.13 shipped, `severity` ignores
|
|
7
7
|
* the model's label and uses a fixed threshold, so the same finding set always
|
|
8
8
|
* yields the same exit code.
|
|
9
9
|
*
|
|
10
10
|
* This is the one place that answers "does this finding block". Every review
|
|
11
|
-
* path
|
|
11
|
+
* path, local, PR, remote and the GitHub App, calls it, so the modes cannot
|
|
12
12
|
* drift apart.
|
|
13
13
|
*/
|
|
14
14
|
export type ReviewBlocking = "model" | "severity";
|
|
@@ -17,7 +17,7 @@ export declare const DEFAULT_REVIEW_BLOCKING: ReviewBlocking;
|
|
|
17
17
|
* so an unset or mistyped value never silently tightens the rule. */
|
|
18
18
|
export declare function reviewBlockingFrom(value: unknown): ReviewBlocking;
|
|
19
19
|
/** One finding, in the shape any of the review paths happens to have. Only
|
|
20
|
-
* `severity` is always available
|
|
20
|
+
* `severity` is always available. The other two describe the model's own
|
|
21
21
|
* claim, which `severity` mode ignores. */
|
|
22
22
|
export type BlockingInput = {
|
|
23
23
|
severity?: string;
|
|
@@ -3,12 +3,12 @@
|
|
|
3
3
|
* The same pull request was reviewed twice, the same P2 finding came back
|
|
4
4
|
* `non-blocking` the first time and `blocking` the second, and the exit code
|
|
5
5
|
* changed because the model's own label moved. `model` keeps that behavior
|
|
6
|
-
* (the model decides) because it is what 0.4.13 shipped
|
|
6
|
+
* (the model decides) because it is what 0.4.13 shipped, `severity` ignores
|
|
7
7
|
* the model's label and uses a fixed threshold, so the same finding set always
|
|
8
8
|
* yields the same exit code.
|
|
9
9
|
*
|
|
10
10
|
* This is the one place that answers "does this finding block". Every review
|
|
11
|
-
* path
|
|
11
|
+
* path, local, PR, remote and the GitHub App, calls it, so the modes cannot
|
|
12
12
|
* drift apart.
|
|
13
13
|
*/
|
|
14
14
|
export const DEFAULT_REVIEW_BLOCKING = "model";
|
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
import type { ParsedFinding } from "../pr/findings.ts";
|
|
2
|
+
import type { ResolvedReference } from "./references.ts";
|
|
2
3
|
import { type Revision, type RevisionFile } from "./revision.ts";
|
|
3
4
|
export type StoredFinding = {
|
|
4
5
|
id: string;
|
|
@@ -12,6 +13,8 @@ export type StoredFinding = {
|
|
|
12
13
|
firstSeenReviewId: string | null;
|
|
13
14
|
/** What maintainers wrote on the finding's inline thread, oldest first. */
|
|
14
15
|
replies?: string[];
|
|
16
|
+
/** What the finding rested on when it was first found. */
|
|
17
|
+
references?: ResolvedReference[];
|
|
15
18
|
};
|
|
16
19
|
export type CarryPrevious = {
|
|
17
20
|
files: RevisionFile[];
|
|
@@ -41,6 +44,11 @@ export type ResolvedFinding = {
|
|
|
41
44
|
severity: string;
|
|
42
45
|
carriedFromId: string | null;
|
|
43
46
|
firstSeenReviewId: string | null;
|
|
47
|
+
/** A carried finding keeps what it rested on. */
|
|
48
|
+
references?: ResolvedReference[];
|
|
49
|
+
/** A new or rewritten finding: the ids the answer named, resolved by the
|
|
50
|
+
* caller against `RULES.json`. */
|
|
51
|
+
referenceIds?: string[];
|
|
44
52
|
};
|
|
45
53
|
export declare function anchorTextFromPatch(patch: string, from: number, to: number): string | null;
|
|
46
54
|
export declare function relocate(anchorText: string | null, oldFrom: number, patch: string): {
|
|
@@ -48,7 +56,7 @@ export declare function relocate(anchorText: string | null, oldFrom: number, pat
|
|
|
48
56
|
to: number;
|
|
49
57
|
} | null;
|
|
50
58
|
/** A guide rebuilt after the previous review invalidates that
|
|
51
|
-
* review's verdicts, its "unchanged" suppression and its open findings
|
|
59
|
+
* review's verdicts, its "unchanged" suppression and its open findings, they
|
|
52
60
|
* were all judged against the old rules. Such a run starts fresh instead:
|
|
53
61
|
* every file is scanned again and old findings are no longer carried over.
|
|
54
62
|
*
|
|
@@ -71,7 +71,7 @@ function guideChanged(previous, currentGuideBuiltAt) {
|
|
|
71
71
|
return (previous.guideBuiltAt ?? "") !== (currentGuideBuiltAt ?? "");
|
|
72
72
|
}
|
|
73
73
|
/** A guide rebuilt after the previous review invalidates that
|
|
74
|
-
* review's verdicts, its "unchanged" suppression and its open findings
|
|
74
|
+
* review's verdicts, its "unchanged" suppression and its open findings, they
|
|
75
75
|
* were all judged against the old rules. Such a run starts fresh instead:
|
|
76
76
|
* every file is scanned again and old findings are no longer carried over.
|
|
77
77
|
*
|
|
@@ -209,7 +209,7 @@ export function buildCarryPromptSection(items, current) {
|
|
|
209
209
|
" not. Report every decision in a final section headed",
|
|
210
210
|
' exactly "## Previous findings", one line per id, exactly as',
|
|
211
211
|
' "- F1: open path:from-to" or "- F1: closed".',
|
|
212
|
-
'Do not repeat an open previous finding under "## Findings"
|
|
212
|
+
'Do not repeat an open previous finding under "## Findings". Report there',
|
|
213
213
|
"only findings from job 1 that are not already covered below.",
|
|
214
214
|
"",
|
|
215
215
|
];
|
|
@@ -280,6 +280,7 @@ export function resolveCarryOutcomes(items, current, visiblePaths, verdicts, par
|
|
|
280
280
|
anchorText: base.anchorText,
|
|
281
281
|
severity: base.severity,
|
|
282
282
|
carriedFromId: base.id,
|
|
283
|
+
...(base.references ? { references: base.references } : {}),
|
|
283
284
|
firstSeenReviewId: base.firstSeenReviewId,
|
|
284
285
|
});
|
|
285
286
|
continue;
|
|
@@ -297,6 +298,7 @@ export function resolveCarryOutcomes(items, current, visiblePaths, verdicts, par
|
|
|
297
298
|
anchorText: base.anchorText,
|
|
298
299
|
severity: base.severity,
|
|
299
300
|
carriedFromId: base.id,
|
|
301
|
+
...(base.references ? { references: base.references } : {}),
|
|
300
302
|
firstSeenReviewId: base.firstSeenReviewId,
|
|
301
303
|
});
|
|
302
304
|
continue;
|
|
@@ -327,6 +329,7 @@ export function resolveCarryOutcomes(items, current, visiblePaths, verdicts, par
|
|
|
327
329
|
: base.anchorText,
|
|
328
330
|
severity: base.severity,
|
|
329
331
|
carriedFromId: base.id,
|
|
332
|
+
...(base.references ? { references: base.references } : {}),
|
|
330
333
|
firstSeenReviewId: base.firstSeenReviewId,
|
|
331
334
|
};
|
|
332
335
|
resolved.push(row);
|
|
@@ -353,6 +356,7 @@ export function resolveCarryOutcomes(items, current, visiblePaths, verdicts, par
|
|
|
353
356
|
anchorText: anchor,
|
|
354
357
|
severity: base.severity,
|
|
355
358
|
carriedFromId: base.id,
|
|
359
|
+
...(base.references ? { references: base.references } : {}),
|
|
356
360
|
firstSeenReviewId: base.firstSeenReviewId,
|
|
357
361
|
};
|
|
358
362
|
resolved.push(row);
|
|
@@ -367,6 +371,9 @@ export function resolveCarryOutcomes(items, current, visiblePaths, verdicts, par
|
|
|
367
371
|
overlap.title = finding.heading || finding.path;
|
|
368
372
|
overlap.lineFrom = finding.from;
|
|
369
373
|
overlap.lineTo = finding.to;
|
|
374
|
+
// The rewritten text rests on what the new answer named.
|
|
375
|
+
overlap.references = undefined;
|
|
376
|
+
overlap.referenceIds = finding.references ?? [];
|
|
370
377
|
continue;
|
|
371
378
|
}
|
|
372
379
|
resolved.push({
|
|
@@ -381,6 +388,7 @@ export function resolveCarryOutcomes(items, current, visiblePaths, verdicts, par
|
|
|
381
388
|
severity: finding.severity ?? "P2",
|
|
382
389
|
carriedFromId: null,
|
|
383
390
|
firstSeenReviewId: null,
|
|
391
|
+
referenceIds: finding.references ?? [],
|
|
384
392
|
});
|
|
385
393
|
}
|
|
386
394
|
return resolved;
|
|
@@ -62,6 +62,7 @@ export function plantumlServerProblem(value) {
|
|
|
62
62
|
// The link is posted as Markdown, so a path that can end it early or be
|
|
63
63
|
// rewritten as prose would break every diagram.
|
|
64
64
|
if (/\s/.test(value) || /[()[\]<>`\\;"]/.test(url.pathname)) {
|
|
65
|
+
// punctuation: code, the message names the semicolon character the path check rejects
|
|
65
66
|
return 'must not contain spaces or any of ( ) [ ] < > ` \\ ; " in its path';
|
|
66
67
|
}
|
|
67
68
|
return null;
|
|
@@ -253,13 +254,13 @@ const PLANTUML_GUIDANCE = `PlantUML selection and minimal syntax:
|
|
|
253
254
|
Write standard PlantUML source in a fenced code block tagged plantuml, from
|
|
254
255
|
@startuml to @enduml. Do not write an image, an image link, a server URL, or
|
|
255
256
|
any other diagram language. Our code decides how the source is shown.
|
|
256
|
-
sequence diagram = ordered calls, webhooks, retries, responses, and timing
|
|
257
|
-
activity diagram = decisions, branches, pipelines, and fallback paths
|
|
257
|
+
sequence diagram = ordered calls, webhooks, retries, responses, and timing.
|
|
258
|
+
activity diagram = decisions, branches, pipelines, and fallback paths.
|
|
258
259
|
class diagram = classes, interfaces, types, inheritance, composition, and
|
|
259
|
-
database entities
|
|
260
|
-
state diagram = lifecycle and state transitions
|
|
261
|
-
component diagram = services, modules, boundaries, and integrations
|
|
262
|
-
deployment diagram = containers, storage, and deployment topology
|
|
260
|
+
database entities.
|
|
261
|
+
state diagram = lifecycle and state transitions.
|
|
262
|
+
component diagram = services, modules, boundaries, and integrations.
|
|
263
|
+
deployment diagram = containers, storage, and deployment topology.
|
|
263
264
|
use case diagram = actors and system capabilities.
|
|
264
265
|
Use the smallest type that matches the evidence. Keep labels short.`;
|
|
265
266
|
/** The system prompt's diagram section for a PlantUML runtime, or `undefined`
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
/** Review prompt, tool loop, and improvement passes
|
|
1
|
+
/** Review prompt, tool loop, and improvement passes, PR adapter calls this
|
|
2
2
|
* after building a `Revision` and carry-over input. */
|
|
3
3
|
import type { GitHubClient, Options } from "../types.ts";
|
|
4
4
|
import type { Snapshot } from "../pr/snapshot.ts";
|
|
@@ -56,7 +56,7 @@ export async function loadGuides(repo) {
|
|
|
56
56
|
guideBuiltAt = getRepo(repo)?.knowledge_built_at ?? null;
|
|
57
57
|
}
|
|
58
58
|
catch {
|
|
59
|
-
// Local CLI review without `serve
|
|
59
|
+
// Local CLI review without `serve`, guides still load from disk.
|
|
60
60
|
}
|
|
61
61
|
// With no `repos` row, the files are the only evidence of when the guide
|
|
62
62
|
// was built. Without this a local review always says `guide unknown` and
|
|
@@ -0,0 +1,68 @@
|
|
|
1
|
+
/** What a review finding rests on.
|
|
2
|
+
*
|
|
3
|
+
* The review answer names ids only: `R12` and `I3` from the guide and
|
|
4
|
+
* instruction lines, `codebase:<section>` for a CODEBASE CONVENTIONS section,
|
|
5
|
+
* and `system` for the review instructions themselves. No pull request number
|
|
6
|
+
* or date ever enters the prompt. Code turns each id into a reference from
|
|
7
|
+
* `RULES.json` at review time, and the result is stored with the finding, so
|
|
8
|
+
* a later `sync` never changes what an old review cited. */
|
|
9
|
+
import { type RulesFile } from "../knowledge/rules.ts";
|
|
10
|
+
/** What references resolve against, from the guides a review was given. */
|
|
11
|
+
export type ReferenceContext = {
|
|
12
|
+
rules: RulesFile | null;
|
|
13
|
+
codebase: string;
|
|
14
|
+
};
|
|
15
|
+
export type ReferenceSource = {
|
|
16
|
+
pr: number;
|
|
17
|
+
url?: string;
|
|
18
|
+
mergedAt?: string;
|
|
19
|
+
} | {
|
|
20
|
+
path: string;
|
|
21
|
+
line: number;
|
|
22
|
+
url?: string;
|
|
23
|
+
};
|
|
24
|
+
export type ResolvedReference = {
|
|
25
|
+
kind: "rule";
|
|
26
|
+
id: string;
|
|
27
|
+
sources: ReferenceSource[];
|
|
28
|
+
} | {
|
|
29
|
+
kind: "instruction";
|
|
30
|
+
id: string;
|
|
31
|
+
sources: ReferenceSource[];
|
|
32
|
+
} | {
|
|
33
|
+
kind: "codebase";
|
|
34
|
+
section?: string;
|
|
35
|
+
} | {
|
|
36
|
+
kind: "system";
|
|
37
|
+
};
|
|
38
|
+
/** The ids as the model wrote them, one per entry, trimmed. */
|
|
39
|
+
export declare function referenceIds(value: unknown): string[];
|
|
40
|
+
/** Turns the ids of one finding into references. An id that is not in
|
|
41
|
+
* `RULES.json` is dropped, so a made-up id never shows. A finding that names
|
|
42
|
+
* nothing usable rests on the review instructions: `system`. */
|
|
43
|
+
export declare function resolveReferences(ids: string[], input: {
|
|
44
|
+
repo: string;
|
|
45
|
+
rules: RulesFile | null;
|
|
46
|
+
codebase: string;
|
|
47
|
+
}): ResolvedReference[];
|
|
48
|
+
/** One reference as a reader sees it:
|
|
49
|
+
* `Rules-12 · #212, #340 · 2026-03`, `Instructions-3 · AGENTS.md:14`,
|
|
50
|
+
* `System Prompt`, `Codebase · Tests`. */
|
|
51
|
+
export declare function referenceLine(reference: ResolvedReference, links?: boolean): string;
|
|
52
|
+
/** The collapsed block under a finding, or `""` when there is nothing. */
|
|
53
|
+
export declare function referencesBlock(references: ResolvedReference[]): string;
|
|
54
|
+
/** Reads a stored `references_json` value. Anything that does not fit is
|
|
55
|
+
* skipped, so a bad row shows no references rather than failing a page. */
|
|
56
|
+
export declare function parseStoredReferences(text: string | null | undefined): ResolvedReference[];
|
|
57
|
+
/** The `references_json` to store for a finding: a carried one keeps what it
|
|
58
|
+
* rested on, a new one resolves the ids its answer named. */
|
|
59
|
+
export declare function referencesJson(finding: {
|
|
60
|
+
references?: ResolvedReference[];
|
|
61
|
+
referenceIds?: string[];
|
|
62
|
+
}, repo: string, context: ReferenceContext | undefined): string;
|
|
63
|
+
/** Gives every row that has no references yet the ones its ids resolve to,
|
|
64
|
+
* so every output (stored row, JSON, carry file) carries the same list. */
|
|
65
|
+
export declare function withResolvedReferences<T extends {
|
|
66
|
+
references?: ResolvedReference[];
|
|
67
|
+
referenceIds?: string[];
|
|
68
|
+
}>(rows: T[], repo: string, context: ReferenceContext | undefined): T[];
|