Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

feat: agent co-author detection + derived decision scope (#803)

Swati Ahuja23f0e663e0fb
diff --git a/packages/api-client/src/types/decisions.ts b/packages/api-client/src/types/decisions.ts
index 93de970a5..26db63460 100644
--- a/packages/api-client/src/types/decisions.ts
+++ b/packages/api-client/src/types/decisions.ts
@@ -25,6 +25,11 @@ export interface DecisionRecordResponse {
last_code_change: string | null;
/** Trust tier of the decision's primary supporting evidence. */
verification?: DecisionVerification;
+ /**
+ * Derived granularity of the decision's blast area. Absent on older
+ * backends; null when the record has no code linkage at all.
+ */
+ scope?: "file" | "module" | "cross-module" | null;
created_at: string;
updated_at: string;
/** Evidence rows backing the record. Populated by the list endpoint only. */
diff --git a/packages/core/src/repowise/core/analysis/decisions/scope.py b/packages/core/src/repowise/core/analysis/decisions/scope.py
new file mode 100644
index 000000000..1e931ee85
--- /dev/null
+++ b/packages/core/src/repowise/core/analysis/decisions/scope.py
@@ -0,0 +1,51 @@
+"""Derived decision granularity — how much of the codebase a decision spans.
+
+Pure derivation from a record's existing linkage fields (no LLM, no I/O), so
+it can run at serialization time for any record, old or new. Levels, narrowest
+first: ``file`` < ``module`` < ``cross-module``. ``None`` means the record has
+no code linkage at all — better no claim than defaulting the least-grounded
+records to the widest level.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+__all__ = ["derive_decision_scope"]
+
+
+def derive_decision_scope(
+ affected_files: Sequence[str] | None,
+ affected_modules: Sequence[str] | None,
+ *,
+ evidence_file: str | None = None,
+) -> str | None:
+ """Return the scope level for one decision record, or ``None``.
+
+ Rules, in order:
+
+ - nothing linked at all: ``file`` when an *evidence_file* pins the record
+ to one file, else ``None`` — explicit file/module linkage always
+ outranks the evidence-file fallback
+ - exactly one affected file → ``file``
+ - otherwise count distinct modules — the explicitly linked ones, or when
+ none are linked, the top-level directories of the affected files: one →
+ ``module``, several → ``cross-module``; multiple root-level files with
+ no directory stay ``file``.
+
+ An evidence line narrows nothing on its own — a line without a resolved
+ symbol still only proves file-level scope — so it is deliberately not a
+ parameter.
+ """
+ files = {f for f in (affected_files or []) if f}
+ modules = {m for m in (affected_modules or []) if m}
+
+ if not files and not modules:
+ return "file" if evidence_file else None
+ if len(files) == 1:
+ return "file"
+ if not modules:
+ modules = {f.split("/", 1)[0] for f in files if "/" in f}
+ if not modules:
+ return "file"
+ return "module" if len(modules) == 1 else "cross-module"
diff --git a/packages/core/src/repowise/core/ingestion/git_indexer/agent_provenance.py b/packages/core/src/repowise/core/ingestion/git_indexer/agent_provenance.py
index 135d8b678..67c0dc6dc 100644
--- a/packages/core/src/repowise/core/ingestion/git_indexer/agent_provenance.py
+++ b/packages/core/src/repowise/core/ingestion/git_indexer/agent_provenance.py
@@ -118,6 +118,23 @@ _AGENT_ALIASES: list[tuple[str, str]] = [
_SERVICE_EMAILS: dict[str, tuple[str, int]] = {
"[email protected]": ("cursor", 1),
"devin-ai-integration[bot]@users.noreply.github.com": ("devin", 1),
+ # Primarily Claude's co-author identity; as AUTHOR it is tier 1 like every
+ # other service e-mail here.
+ "[email protected]": ("claude", 1),
+}
+
+# Agent-vendor e-mail domains → the exact agent identities each vendor emits.
+# A domain alone is never enough — real humans work at these companies — so an
+# identity here counts only when the LOCAL PART fully matches the vendor's
+# allowlisted agent identity (anchored fullmatch, never a substring scan, and
+# per vendor, never a flat token set): ``[email protected]`` and the
+# ``claude-<model>`` slugs match; ``[email protected]``,
+# ``[email protected]`` and ``[email protected]`` never do.
+_VENDOR_DOMAIN_LOCALS: dict[str, tuple[re.Pattern[str], str]] = {
+ "anthropic.com": (re.compile(r"claude(?:-(?:opus|sonnet|haiku)(?:-[\w.]+)?)?"), "claude"),
+ "openai.com": (re.compile(r"codex"), "codex"),
+ "cursor.com": (re.compile(r"cursor(?:agent)?"), "cursor"),
+ "devin.ai": (re.compile(r"devin"), "devin"),
}
# Commit-message footers → agent (exact service phrases only).
@@ -158,6 +175,13 @@ _COAUTHOR_PATTERNS: list[tuple[str, str]] = [
_AIDER_NAME_RE = re.compile(r"\(aider\)\s*$")
+# Any ``Co-authored-by:`` trailer's e-mail, for registry-based identity
+# matching: agent service identities the explicit patterns above don't name
+# (bot noreply logins, exact service e-mails, vendor-domain identities) are
+# resolved through the same :meth:`AgentProvenanceClassifier._identity_match`
+# registry — one list, no parallel patterns.
+_COAUTHOR_EMAIL_RE = re.compile(r"^co-authored-by:[^<\n]*<([^>\s]+)>", re.IGNORECASE | re.MULTILINE)
+
def _normalize_agent_token(value: str, *, allow_slug: bool) -> str | None:
"""Resolve a free-text tool/agent name to a canonical label.
@@ -274,6 +298,10 @@ class AgentProvenanceClassifier:
m = self._bot_email_re.match(e)
if m:
return (_BOT_LOGINS[m.group(1).lower()], 1)
+ local, _, domain = e.partition("@")
+ vendor = _VENDOR_DOMAIN_LOCALS.get(domain)
+ if vendor and vendor[0].fullmatch(local):
+ return (vendor[1], 1)
return None
def classify(
@@ -328,6 +356,10 @@ class AgentProvenanceClassifier:
for pat, agent in self._coauthors:
if pat.search(message):
return AgentProvenance(agent, 3, "coauthor_trailer", "high")
+ for m in _COAUTHOR_EMAIL_RE.finditer(message):
+ hit = self._identity_match(m.group(1))
+ if hit:
+ return AgentProvenance(hit[0], 3, "coauthor_trailer", "high")
return _HUMAN
diff --git a/packages/server/src/repowise/server/schemas/decisions.py b/packages/server/src/repowise/server/schemas/decisions.py
index 76b363d17..e7c4c48f8 100644
--- a/packages/server/src/repowise/server/schemas/decisions.py
+++ b/packages/server/src/repowise/server/schemas/decisions.py
@@ -7,6 +7,8 @@ from datetime import datetime
from pydantic import BaseModel
+from repowise.core.analysis.decisions.scope import derive_decision_scope
+
class EvidencePreview(BaseModel):
"""The top-ranked evidence row, slimmed for list rows."""
@@ -38,6 +40,10 @@ class DecisionRecordResponse(BaseModel):
confidence: float
staleness_score: float
verification: str = "unverified"
+ # Derived granularity: file | module | cross-module, or None when the
+ # record has no code linkage at all. Computed at serialization time from
+ # the linkage fields, so old records get it too.
+ scope: str | None = None
superseded_by: str | None
last_code_change: datetime | None
created_at: datetime
@@ -51,6 +57,8 @@ class DecisionRecordResponse(BaseModel):
@classmethod
def from_orm(cls, obj: object) -> DecisionRecordResponse:
+ affected_files = json.loads(obj.affected_files_json) # type: ignore[attr-defined]
+ affected_modules = json.loads(obj.affected_modules_json) # type: ignore[attr-defined]
return cls(
id=obj.id, # type: ignore[attr-defined]
repository_id=obj.repository_id, # type: ignore[attr-defined]
@@ -67,8 +75,8 @@ class DecisionRecordResponse(BaseModel):
rationale=obj.rationale, # type: ignore[attr-defined]
alternatives=json.loads(obj.alternatives_json), # type: ignore[attr-defined]
consequences=json.loads(obj.consequences_json), # type: ignore[attr-defined]
- affected_files=json.loads(obj.affected_files_json), # type: ignore[attr-defined]
- affected_modules=json.loads(obj.affected_modules_json), # type: ignore[attr-defined]
+ affected_files=affected_files,
+ affected_modules=affected_modules,
tags=json.loads(obj.tags_json), # type: ignore[attr-defined]
source=obj.source, # type: ignore[attr-defined]
evidence_commits=json.loads(obj.evidence_commits_json), # type: ignore[attr-defined]
@@ -77,6 +85,11 @@ class DecisionRecordResponse(BaseModel):
confidence=obj.confidence, # type: ignore[attr-defined]
staleness_score=obj.staleness_score, # type: ignore[attr-defined]
verification=obj.verification, # type: ignore[attr-defined]
+ scope=derive_decision_scope(
+ affected_files,
+ affected_modules,
+ evidence_file=obj.evidence_file, # type: ignore[attr-defined]
+ ),
superseded_by=obj.superseded_by, # type: ignore[attr-defined]
last_code_change=obj.last_code_change, # type: ignore[attr-defined]
created_at=obj.created_at, # type: ignore[attr-defined]
diff --git a/packages/types/src/decisions.ts b/packages/types/src/decisions.ts
index 2af5ec983..1d098b941 100644
--- a/packages/types/src/decisions.ts
+++ b/packages/types/src/decisions.ts
@@ -19,6 +19,9 @@ export type DecisionSource =
| "readme_mining"
| "cli";
+/** Derived granularity of a decision's blast area, narrowest first. */
+export type DecisionScope = "file" | "module" | "cross-module";
+
export interface DecisionRecord {
id: string;
repository_id: string;
@@ -42,6 +45,11 @@ export interface DecisionRecord {
last_code_change: string | null;
/** Trust tier of the decision's primary supporting evidence. Optional for back-compat. */
verification?: DecisionVerification;
+ /**
+ * Derived granularity level. Optional for back-compat with older backends;
+ * null when the record has no code linkage at all.
+ */
+ scope?: DecisionScope | null;
created_at: string;
updated_at: string;
/** Number of evidence rows backing the record. List endpoint only. */
diff --git a/packages/ui/__tests__/decisions/decisions-table.test.tsx b/packages/ui/__tests__/decisions/decisions-table.test.tsx
index 3fc593a7c..fc6ee9f67 100644
--- a/packages/ui/__tests__/decisions/decisions-table.test.tsx
+++ b/packages/ui/__tests__/decisions/decisions-table.test.tsx
@@ -66,6 +66,74 @@ describe("DecisionsTable", () => {
});
});
+ it("renders a scope badge when the record carries one", () => {
+ render(
+ <DecisionsTable
+ {...baseProps}
+ decisions={[
+ makeDecision({ id: "1", scope: "file" }),
+ makeDecision({ id: "2", title: "Adopt SWR", scope: "cross-module" }),
+ ]}
+ />,
+ );
+ // The scope <select> options share these labels, so exclude them.
+ expect(screen.getAllByText("File").some((el) => el.tagName !== "OPTION")).toBe(true);
+ expect(
+ screen.getAllByText("Cross-module").some((el) => el.tagName !== "OPTION"),
+ ).toBe(true);
+ });
+
+ it("invokes onFiltersChange when the scope filter changes", () => {
+ const onFiltersChange = vi.fn();
+ render(
+ <DecisionsTable
+ {...baseProps}
+ onFiltersChange={onFiltersChange}
+ decisions={[makeDecision({ id: "1", scope: "file" })]}
+ />,
+ );
+ fireEvent.change(screen.getByLabelText("Filter by scope"), {
+ target: { value: "module" },
+ });
+ expect(onFiltersChange).toHaveBeenCalledWith({
+ status: "all",
+ source: "all",
+ scope: "module",
+ });
+ });
+
+ it("hides the scope filter and ignores a scope value when no record carries scope", () => {
+ render(
+ <DecisionsTable
+ {...baseProps}
+ filters={{ status: "all", source: "all", scope: "file" }}
+ decisions={[
+ makeDecision({ id: "1", title: "Pick Postgres" }),
+ makeDecision({ id: "2", title: "Adopt SWR" }),
+ ]}
+ />,
+ );
+ expect(screen.queryByLabelText("Filter by scope")).not.toBeInTheDocument();
+ // Rows are NOT filtered out by the stale scope value.
+ expect(screen.getByText("Pick Postgres")).toBeInTheDocument();
+ expect(screen.getByText("Adopt SWR")).toBeInTheDocument();
+ });
+
+ it("filters rows client-side by scope", () => {
+ render(
+ <DecisionsTable
+ {...baseProps}
+ filters={{ status: "all", source: "all", scope: "file" }}
+ decisions={[
+ makeDecision({ id: "1", title: "Pick Postgres", scope: "file" }),
+ makeDecision({ id: "2", title: "Adopt SWR", scope: "cross-module" }),
+ ]}
+ />,
+ );
+ expect(screen.getByText("Pick Postgres")).toBeInTheDocument();
+ expect(screen.queryByText("Adopt SWR")).not.toBeInTheDocument();
+ });
+
it("renders a retry button when an error is supplied with no decisions", () => {
const onRetry = vi.fn();
render(
diff --git a/packages/ui/src/decisions/decisions-table.tsx b/packages/ui/src/decisions/decisions-table.tsx
index 24945a9dc..c8db10480 100644
--- a/packages/ui/src/decisions/decisions-table.tsx
+++ b/packages/ui/src/decisions/decisions-table.tsx
@@ -14,6 +14,7 @@ import type {
DecisionRecord,
DecisionStatus,
DecisionSource,
+ DecisionScope,
} from "@repowise-dev/types/decisions";
const STATUS_VARIANT: Record<string, "default" | "fresh" | "stale" | "outdated" | "outline" | "accent"> = {
@@ -30,12 +31,25 @@ const SOURCE_LABEL: Record<string, string> = {
cli: "Manual",
};
+const SCOPE_LABEL: Record<string, string> = {
+ file: "File",
+ module: "Module",
+ "cross-module": "Cross-module",
+};
+
export type DecisionStatusFilter = DecisionStatus | "all";
export type DecisionSourceFilter = DecisionSource | "all";
+export type DecisionScopeFilter = DecisionScope | "all";
export interface DecisionsTableFilters {
status: DecisionStatusFilter;
source: DecisionSourceFilter;
+ /**
+ * Optional for back-compat with callers that predate scope. Unlike
+ * status/source (server query params), scope is derived at serialization
+ * time, so the table filters rows client-side.
+ */
+ scope?: DecisionScopeFilter;
}
export interface DecisionsTableProps {
@@ -71,6 +85,16 @@ export function DecisionsTable({
const prefix = linkPrefix ?? `/repos/${repoId}`;
const Link = LinkComponent;
+ // Scope is derived at serialization time (no server-side query param), so
+ // this filter applies client-side to the fetched rows. Backends that don't
+ // serve scope yet leave every row null: hide the filter (it could only
+ // empty the table) and ignore any lingering scope value.
+ const scopeFilter = filters.scope ?? "all";
+ const hasScope = (decisions ?? []).some((d) => d.scope != null);
+ const visibleDecisions = (decisions ?? []).filter(
+ (d) => !hasScope || scopeFilter === "all" || d.scope === scopeFilter,
+ );
+
const columns: ResponsiveColumn<DecisionRecord>[] = [
{
key: "title",
@@ -123,6 +147,17 @@ export function DecisionsTable({
cellClassName: "text-[var(--color-text-secondary)]",
render: (d) => SOURCE_LABEL[d.source] ?? d.source,
},
+ {
+ key: "scope",
+ header: "Scope",
+ priority: 3,
+ render: (d) =>
+ d.scope ? (
+ <Badge variant="outline">{SCOPE_LABEL[d.scope] ?? d.scope}</Badge>
+ ) : (
+ <span className="text-[var(--color-text-tertiary)]">—</span>
+ ),
+ },
{
key: "trust",
header: "Trust",
@@ -234,11 +269,26 @@ export function DecisionsTable({
<option value="readme_mining">Docs mining</option>
<option value="cli">Manual</option>
</select>
+ {hasScope && (
+ <select
+ value={filters.scope ?? "all"}
+ onChange={(e) =>
+ onFiltersChange({ ...filters, scope: e.target.value as DecisionScopeFilter })
+ }
+ aria-label="Filter by scope"
+ className="w-full sm:w-auto rounded-md border border-[var(--color-border-default)] bg-[var(--color-bg-surface)] px-3 py-1.5 text-sm text-[var(--color-text-primary)]"
+ >
+ <option value="all">All scopes</option>
+ <option value="file">File</option>
+ <option value="module">Module</option>
+ <option value="cross-module">Cross-module</option>
+ </select>
+ )}
</div>
<ResponsiveTable
columns={columns}
- rows={decisions ?? []}
+ rows={visibleDecisions}
rowKey={(d) => d.id}
empty={empty}
/>
diff --git a/tests/unit/analysis/test_decision_scope.py b/tests/unit/analysis/test_decision_scope.py
new file mode 100644
index 000000000..4b9e90b8b
--- /dev/null
+++ b/tests/unit/analysis/test_decision_scope.py
@@ -0,0 +1,55 @@
+"""Unit tests for the derived decision scope level — one test per rule branch."""
+
+from __future__ import annotations
+
+from repowise.core.analysis.decisions.scope import derive_decision_scope
+
+
+def test_single_file_is_file() -> None:
+ assert derive_decision_scope(["a/b.py"], []) == "file"
+
+
+def test_evidence_file_counts_when_nothing_else_is_linked() -> None:
+ assert derive_decision_scope([], [], evidence_file="a/b.py") == "file"
+
+
+def test_modules_outrank_the_evidence_file_fallback() -> None:
+ assert (
+ derive_decision_scope([], ["server", "ui", "core"], evidence_file="README.md")
+ == "cross-module"
+ )
+ assert derive_decision_scope([], ["server"], evidence_file="README.md") == "module"
+
+
+def test_multiple_files_one_module_is_module() -> None:
+ assert derive_decision_scope(["a/b.py", "a/c.py"], ["a"]) == "module"
+
+
+def test_multiple_files_without_modules_infers_from_top_level_dirs() -> None:
+ assert derive_decision_scope(["a/b.py", "a/c.py"], []) == "module"
+ assert derive_decision_scope(["a/b.py", "z/c.py"], []) == "cross-module"
+
+
+def test_multiple_root_level_files_without_modules_is_file() -> None:
+ assert derive_decision_scope(["README.md", "LICENSE"], []) == "file"
+
+
+def test_files_spanning_modules_is_cross_module() -> None:
+ assert derive_decision_scope(["a/b.py", "z/c.py"], ["a", "z"]) == "cross-module"
+
+
+def test_only_one_module_is_module() -> None:
+ assert derive_decision_scope([], ["a"]) == "module"
+
+
+def test_only_multiple_modules_is_cross_module() -> None:
+ assert derive_decision_scope([], ["a", "z"]) == "cross-module"
+
+
+def test_nothing_linked_is_none() -> None:
+ assert derive_decision_scope([], []) is None
+ assert derive_decision_scope(None, None) is None
+
+
+def test_duplicate_file_entries_collapse_to_one_file() -> None:
+ assert derive_decision_scope(["a/b.py", "a/b.py"], []) == "file"
diff --git a/tests/unit/ingestion/test_agent_provenance.py b/tests/unit/ingestion/test_agent_provenance.py
index 64b75e3cf..5e2c0b12d 100644
--- a/tests/unit/ingestion/test_agent_provenance.py
+++ b/tests/unit/ingestion/test_agent_provenance.py
@@ -27,7 +27,9 @@ from repowise.core.ingestion.git_indexer.records import _CommitRec
CLF = AgentProvenanceClassifier()
-def _classify(an="Dev", ae="[email protected]", cn="Dev", ce="[email protected]", msg="feat: thing", note_agent=None):
+def _classify(
+ an="Dev", ae="[email protected]", cn="Dev", ce="[email protected]", msg="feat: thing", note_agent=None
+):
return CLF.classify(an, ae, cn, ce, msg, note_agent=note_agent)
@@ -101,6 +103,45 @@ def test_coauthor_trailer_is_tier3() -> None:
assert (prov.agent, prov.autonomy_tier, prov.channel) == ("claude", 3, "coauthor_trailer")
+def test_coauthor_trailer_with_bot_noreply_email_is_tier3() -> None:
+ """A co-author trailer whose e-mail is a known bot noreply identity is
+ matched through the identity registry — no explicit pattern needed."""
+ msg = (
+ "chore: bump deps\n\n"
+ "Co-authored-by: gemini-code-assist[bot] "
+ "<176961590+gemini-code-assist[bot]@users.noreply.github.com>"
+ )
+ prov = _classify(msg=msg)
+ assert (prov.agent, prov.autonomy_tier, prov.channel) == ("gemini", 3, "coauthor_trailer")
+
+
+def test_coauthor_trailer_with_vendor_domain_email_is_tier3() -> None:
+ msg = "fix: retry\n\nCo-authored-by: Codex <[email protected]>"
+ prov = _classify(msg=msg)
+ assert (prov.agent, prov.autonomy_tier, prov.channel) == ("codex", 3, "coauthor_trailer")
+
+
+def test_agent_local_part_at_vendor_domain_author_is_tier1() -> None:
+ prov = _classify(an="Claude", ae="[email protected]")
+ assert (prov.agent, prov.autonomy_tier, prov.channel) == ("claude", 1, "service_email")
+ prov = _classify(an="Claude", ae="[email protected]")
+ assert (prov.agent, prov.autonomy_tier, prov.channel) == ("claude", 1, "service_email")
+
+
+def test_vendor_domain_committer_over_human_author_is_tier2() -> None:
+ prov = _classify(cn="Claude", ce="[email protected]")
+ assert (prov.agent, prov.autonomy_tier, prov.channel) == ("claude", 2, "service_committer")
+
+
+def test_human_at_vendor_domain_is_unlabelled() -> None:
+ """The vendor domain alone must never label a commit — the local part has
+ to name the agent."""
+ prov = _classify(an="Jane Doe", ae="[email protected]")
+ assert prov.agent is None
+ msg = "fix: typo\n\nCo-authored-by: Jane Doe <[email protected]>"
+ assert _classify(msg=msg).agent is None
+
+
def test_tier_precedence_footer_beats_trailer() -> None:
"""A commit carrying both a T2 footer and a T3 trailer is T2 — first match
in tier order wins."""
@@ -129,6 +170,45 @@ def test_mentioning_claude_in_prose_is_not_a_footer() -> None:
assert prov.agent is None
+def test_vendor_domain_lookalike_locals_are_unlabelled() -> None:
+ """The vendor-domain channel is an exact allowlist on the local part —
+ a human at an agent vendor whose address merely CONTAINS an agent token
+ must never classify as an agent."""
+ for email in (
+ ):
+ assert _classify(an="Someone", ae=email).agent is None, email
+
+
+def test_vendor_domain_allowlists_are_per_vendor_not_a_flat_set() -> None:
+ """codex is OpenAI's agent identity, not Anthropic's."""
+ assert _classify(an="X", ae="[email protected]").agent is None
+
+
+def test_coauthor_human_name_containing_agent_token_is_unlabelled() -> None:
+ msg = "feat: sonata\n\nCo-authored-by: Claude Debussy <[email protected]>"
+ assert _classify(msg=msg).agent is None
+
+
+def test_mixed_case_coauthor_trailer_matches() -> None:
+ msg = "fix: z\n\nCO-AUTHORED-BY: Claude <[email protected]>"
+ prov = _classify(msg=msg)
+ assert (prov.agent, prov.autonomy_tier, prov.channel) == ("claude", 3, "coauthor_trailer")
+
+
+def test_multi_coauthor_human_first_agent_second_is_detected() -> None:
+ msg = (
+ "feat: pairing\n\n"
+ "Co-authored-by: Jane Doe <[email protected]>\n"
+ "Co-authored-by: Claude <[email protected]>"
+ )
+ prov = _classify(msg=msg)
+ assert (prov.agent, prov.autonomy_tier) == ("claude", 3)
+
+
def test_devin_service_trailer_matches() -> None:
msg = (
"feat: connector\n\n"