AI collaboration: recognition ruleset + in-adapter GitHub author classification (foundation) (#121)

provider/github/github.go

@@ -18,6 +18,7 @@ import (
 	"time"
 
 	"github.com/gkanitz/coderepute/provider"
+	"github.com/gkanitz/coderepute/provider/recognition"
 )
 
 const defaultBaseURL = "https://api.github.com"
@@ -208,6 +209,7 @@ type githubFilesResponse struct {
 type apiUser struct {
 	Login string `json:"login"`
 	ID    int64  `json:"id"`
+	Type  string `json:"type,omitempty"`
 }
 
 type apiPull struct {
@@ -259,6 +261,7 @@ func (a *Adapter) FetchActivity(ctx context.Context, opts provider.FetchOptions)
 		}
 	}
 	as.AccessManifest = a.counting.Manifest(provider.GitHubNeverRequested(), "All requests are to the GitHub REST API and GraphQL API. No repository contents, file contents, diffs, branch names, colleague profiles, or commit data are ever requested.")
+	as.AccessManifest.AIRecognitionVersion = recognition.Version()
 	return as, nil
 }
 
@@ -268,6 +271,7 @@ type pendingReview struct {
 	prNumber    int64
 	submittedAt time.Time
 	state       string
+	authorClass string // classification of the reviewed PR's author
 }
 
 // fetchRepoActivity collects one repo's activity into the set. Everything
@@ -318,6 +322,9 @@ func (a *Adapter) fetchRepoActivity(ctx context.Context, repo string, subjectID
 			continue
 		}
 		// Someone else's PR: only the subject's in-window reviews matter.
+		// Classify the PR author and record only the class string --
+		// the colleague's identity never leaves the adapter.
+		authorClass := recognition.Classify(p.User.Login, p.User.Type)
 		for _, rv := range reviews {
 			if rv.User.ID != subjectID || !inWindow(rv.SubmittedAt, window) {
 				continue
@@ -326,6 +333,7 @@ func (a *Adapter) fetchRepoActivity(ctx context.Context, repo string, subjectID
 				prNumber:    p.Number,
 				submittedAt: rv.SubmittedAt,
 				state:       rv.State,
+				authorClass: authorClass,
 			})
 		}
 	}
@@ -343,6 +351,7 @@ func (a *Adapter) fetchRepoActivity(ctx context.Context, repo string, subjectID
 			SubmittedAt:  rv.submittedAt,
 			State:        rv.state,
 			CommentCount: commentCounts[rv.prNumber],
+			AuthorClass:  rv.authorClass,
 		})
 	}
 	return nil

provider/github/github_test.go

@@ -577,3 +577,131 @@ func TestFetchActivityWidenedReviewAllTimeWindow(t *testing.T) {
 		t.Errorf("got %d PRs, want 1", len(as.PullRequests))
 	}
 }
+
+// TestFetchActivityAuthorClassification verifies that the author of a reviewed
+// PR is classified against the recognition ruleset and the class string is
+// recorded on the Review, without leaking the colleague's identity.
+func TestFetchActivityAuthorClassification(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/users/octocat":
+			w.Header().Set("X-OAuth-Scopes", "repo")
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`{"login":"octocat","id":583231}`))
+		case "/repos/acme/widgets/pulls":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[
+				{"number":1,"user":{"login":"copilot[bot]","id":999001,"type":"Bot"},"created_at":"2026-02-15T08:00:00Z","updated_at":"2026-02-20T08:00:00Z"},
+				{"number":2,"user":{"login":"human-colleague","id":888001},"created_at":"2026-02-15T08:00:00Z","updated_at":"2026-02-20T08:00:00Z"}
+			]`))
+		case "/repos/acme/widgets/pulls/1/reviews":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[
+				{"user":{"login":"octocat","id":583231},"state":"APPROVED","submitted_at":"2026-02-20T09:00:00Z"}
+			]`))
+		case "/repos/acme/widgets/pulls/2/reviews":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[
+				{"user":{"login":"octocat","id":583231},"state":"COMMENTED","submitted_at":"2026-02-20T10:00:00Z"}
+			]`))
+		case "/repos/acme/widgets/pulls/comments":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[]`))
+		default:
+			http.Error(w, "not found", http.StatusNotFound)
+		}
+	}))
+	t.Cleanup(srv.Close)
+
+	adapter := github.New("test-token", github.WithBaseURL(srv.URL))
+	as, err := adapter.FetchActivity(context.Background(), provider.FetchOptions{
+		Repos:   []string{"acme/widgets"},
+		Subject: "octocat",
+		Window: provider.Window{
+			Since: time.Date(2026, 1, 1, 0, 0, 0, 0, time.UTC),
+			Until: time.Date(2026, 6, 1, 0, 0, 0, 0, time.UTC),
+		},
+	})
+	if err != nil {
+		t.Fatalf("FetchActivity: %v", err)
+	}
+
+	// Expect two reviews: one on a copilot-authored PR, one on a human-authored PR.
+	if len(as.ReviewsGiven) != 2 {
+		t.Fatalf("got %d reviews given, want 2", len(as.ReviewsGiven))
+	}
+
+	// Review on copilot-authored PR should have AuthorClass = "copilot".
+	var copilotReview, humanReview *provider.Review
+	for i, rv := range as.ReviewsGiven {
+		if rv.AuthorClass == "copilot" {
+			copilotReview = &as.ReviewsGiven[i]
+		} else if rv.AuthorClass == "" {
+			humanReview = &as.ReviewsGiven[i]
+		}
+	}
+
+	if copilotReview == nil {
+		t.Errorf("no review with AuthorClass 'copilot' found; all reviews: %+v", as.ReviewsGiven)
+	}
+	if humanReview == nil {
+		t.Errorf("no review with AuthorClass '' (human) found; all reviews: %+v", as.ReviewsGiven)
+	}
+
+	// Verify that no colleague identity (login, ID) appears in ReviewsGiven.
+	dump := fmt.Sprintf("%+v", as.ReviewsGiven)
+	for _, forbidden := range []string{"copilot[bot]", "999001", "888001"} {
+		if strings.Contains(dump, forbidden) {
+			t.Errorf("Review carries prohibited colleague identity %q", forbidden)
+		}
+	}
+}
+
+// TestFetchActivityBotAuthorClassification verifies that a PR authored by an
+// unknown bot (matched only by type:"Bot" or *[bot] login) gets AuthorClass "bot".
+func TestFetchActivityBotAuthorClassification(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/users/octocat":
+			w.Header().Set("X-OAuth-Scopes", "repo")
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`{"login":"octocat","id":583231}`))
+		case "/repos/acme/widgets/pulls":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[
+				{"number":1,"user":{"login":"some-unknown-tool[bot]","id":999002,"type":"Bot"},"created_at":"2026-02-15T08:00:00Z","updated_at":"2026-02-20T08:00:00Z"}
+			]`))
+		case "/repos/acme/widgets/pulls/1/reviews":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[
+				{"user":{"login":"octocat","id":583231},"state":"APPROVED","submitted_at":"2026-02-20T09:00:00Z"}
+			]`))
+		case "/repos/acme/widgets/pulls/comments":
+			w.Header().Set("Content-Type", "application/json")
+			w.Write([]byte(`[]`))
+		default:
+			http.Error(w, "not found", http.StatusNotFound)
+		}
+	}))
+	t.Cleanup(srv.Close)
+
+	adapter := github.New("test-token", github.WithBaseURL(srv.URL))
+	as, err := adapter.FetchActivity(context.Background(), provider.FetchOptions{
+		Repos:   []string{"acme/widgets"},
+		Subject: "octocat",
+		Window: provider.Window{
+			Since: time.Date(2026, 1, 1, 0, 0, 0, 0, time.UTC),
+			Until: time.Date(2026, 6, 1, 0, 0, 0, 0, time.UTC),
+		},
+	})
+	if err != nil {
+		t.Fatalf("FetchActivity: %v", err)
+	}
+
+	if len(as.ReviewsGiven) != 1 {
+		t.Fatalf("got %d reviews given, want 1", len(as.ReviewsGiven))
+	}
+	if as.ReviewsGiven[0].AuthorClass != "bot" {
+		t.Errorf("AuthorClass = %q, want %q (unknown bot should get 'bot')", as.ReviewsGiven[0].AuthorClass, "bot")
+	}
+}

provider/manifest.go

@@ -27,6 +27,10 @@ type Manifest struct {
 	Endpoints      []EndpointCount `json:"endpoints"`
 	NeverRequested []string        `json:"never_requested"`
 	Notes          string          `json:"notes"`
+	// AIRecognitionVersion is the version of the embedded recognition ruleset
+	// (airuleset.json) used during this fetch, or zero if no classification
+	// was performed.
+	AIRecognitionVersion int `json:"ai_recognition_version,omitempty"`
 }
 
 // RouteEntry maps one URL pattern to its route class. The Pattern is a

provider/provider.go

@@ -84,6 +84,12 @@ type Review struct {
 	// populated from diff-shape data when available. Zero means "unknown /
 	// no diff data" and triggers the fallback deep-review threshold.
 	PRLines int
+	// AuthorClass is the classification of the reviewed PR's author: a
+	// canonical agent id ("copilot", "devin") when the author matches the
+	// curated recognition ruleset, "bot" when matched only by structural
+	// bot-type indicators, or "" for a human author. Populated inside the
+	// adapter; the colleague's identity never leaves the adapter.
+	AuthorClass string
 }
 
 // ReviewComment is a single review comment written or received by the subject.

provider/recognition/airuleset.json

@@ -0,0 +1,45 @@
+{
+  "version": 1,
+  "entries": [
+    {
+      "login": "copilot[bot]",
+      "agent": "copilot"
+    },
+    {
+      "login": "devin[bot]",
+      "agent": "devin"
+    },
+    {
+      "login": "codeium[bot]",
+      "agent": "codeium"
+    },
+    {
+      "login": "tabnine[bot]",
+      "agent": "tabnine"
+    },
+    {
+      "login": "amazon-q-developer[bot]",
+      "agent": "amazon-q"
+    },
+    {
+      "login": "coderabbit[bot]",
+      "agent": "coderabbit"
+    },
+    {
+      "login": "codesee[bot]",
+      "agent": "codesee"
+    },
+    {
+      "login": "github-actions[bot]",
+      "agent": "github-actions"
+    },
+    {
+      "login": "dependabot[bot]",
+      "agent": "dependabot"
+    },
+    {
+      "login": "github-merge-queue[bot]",
+      "agent": "github-merge-queue"
+    }
+  ]
+}

provider/recognition/recognition.go

@@ -0,0 +1,90 @@
+// Package recognition provides an embedded, versioned ruleset for classifying
+// PR/MR authors as human, bot, or a recognized AI agent. The ruleset follows
+// the same go:embed pattern as metrics/bands/bands.json.
+package recognition
+
+import (
+	"embed"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"sync"
+)
+
+//go:embed airuleset.json
+var rulesetFS embed.FS
+
+// Entry maps a known agent login to its canonical agent id.
+type Entry struct {
+	Login string `json:"login"`
+	Agent string `json:"agent"`
+}
+
+// Ruleset is the top-level structure of the embedded airuleset.json.
+type Ruleset struct {
+	Version int     `json:"version"`
+	Entries []Entry `json:"entries"`
+}
+
+var (
+	once     sync.Once
+	loginMap map[string]string // login -> canonical agent id
+	version  int
+)
+
+// load reads and parses the embedded airuleset.json once.
+func load() {
+	raw, err := rulesetFS.ReadFile("airuleset.json")
+	if err != nil {
+		panic(fmt.Sprintf("recognition: embedded airuleset.json: %v", err))
+	}
+	var rs Ruleset
+	if err := json.Unmarshal(raw, &rs); err != nil {
+		panic(fmt.Sprintf("recognition: parse embedded airuleset.json: %v", err))
+	}
+	version = rs.Version
+	loginMap = make(map[string]string, len(rs.Entries))
+	for _, e := range rs.Entries {
+		loginMap[strings.ToLower(e.Login)] = e.Agent
+	}
+}
+
+// agentIDForLogin returns the canonical agent id for a known agent login, or
+// false if the login is not in the curated ruleset.
+func agentIDForLogin(login string) (string, bool) {
+	once.Do(load)
+	aid, ok := loginMap[strings.ToLower(login)]
+	return aid, ok
+}
+
+// Classify returns the canonical agent id when the given login+type matches a
+// curated ruleset entry, "bot" when the match is structural (GitHub type:"Bot"
+// or *[bot] login pattern), or "" for a human author. The login comparison is
+// case-insensitive.
+func Classify(login, userType string) string {
+	if login == "" {
+		return ""
+	}
+	// Layer 1: curated agent ruleset.
+	if aid, ok := agentIDForLogin(login); ok {
+		return aid
+	}
+	// Layer 2: structural bot-type.
+	// GitHub API returns type:"Bot" for bot accounts.
+	if userType == "Bot" {
+		return "bot"
+	}
+	// GitHub bot logins follow the *[bot] pattern (e.g. "copilot[bot]").
+	// This also catches ruleset entries that didn't match above, though
+	// that's redundant — ruleset entries use [bot] logins.
+	if strings.HasSuffix(login, "[bot]") {
+		return "bot"
+	}
+	return ""
+}
+
+// Version returns the embedded airuleset.json version number.
+func Version() int {
+	once.Do(load)
+	return version
+}