feat: let commits-per-repo 0 mean every commit

The pagination guard broke on seen >= maxPerRepo, so a cap of 0 stopped
before the first page and rendered empty productive-time,
productive-weekday and most-commit-language cards. That inverted the
meaning 0 carries for top-repos, where it already means unlimited.

Treat a cap of 0 or less as no cap. Verified against a real repo: a cap
of 100 still stops at 100, while 500 and 0 both walk the full 382-commit
history.

CLAUDE.md

@@ -0,0 +1,52 @@
+# ghstats
+
+Single-binary Go CLI that renders GitHub profile cards as SVG, wrapped by a
+published GitHub Action. `main.go` parses flags → `internal/github` fetches →
+`internal/card` renders → `internal/theme` supplies palettes.
+
+## Adding or changing a config knob
+
+A new flag, input, or default is not done when the code compiles. Every knob
+has a surface in five places, and a change that lands in some but not all of
+them ships a half-wired feature or a README that lies:
+
+1. `main.go` — the CLI flag
+2. `action.yml` — the matching Action input and its default
+3. `entrypoint.sh` — the `INPUT_*` → flag translation
+4. `README.md` — **both** reference tables (Action inputs, CLI flags) **and
+   the runnable examples**: the workflow YAML under "Use as a GitHub Action"
+   and the `ghstats …` command under "Use as a CLI"
+5. `docs/system-architecture.md` — when the knob changes the fetch pipeline
+   or its query cost
+
+The examples are the step most often missed, and they are what people copy.
+Grep the flag name across the repo before calling the change complete; every
+hit that is a reference table or example should mention it.
+
+## Behavior changes are user-visible
+
+The Action is published to the Marketplace, so existing callers get whatever
+`@v1` points at. Default-off for anything that moves numbers on already-
+rendered cards, and say so in the input description. Releases are tag-driven:
+push `vX.Y.Z`, and `release.yml` tests, builds, and force-moves the floating
+`v1` tag.
+
+## Cards and the demo gallery
+
+Adding a card means updating `internal/card/card.go`, the card table in
+`README.md`, and the layout in `.github/workflows/demo.yml` — the demo
+generator embeds a hardcoded list of SVGs, so a new card renders into every
+theme directory but stays invisible in the gallery until it is added there.
+Per-theme pages mirror the table layout in the author's profile README at
+`tiennm99/tiennm99`.
+
+## GitHub API constraints worth remembering
+
+- `commitContributionsByRepository` caps at 100 repos per query and truncates
+  silently. Contribution years are queried by quarter, and a saturated quarter
+  is re-asked by month. Only widen windows with that ceiling in mind.
+- `contributionCalendar` is clipped to the exact `from`/`to`, with no
+  week-boundary spillover, so disjoint windows never double-count days.
+- Test files named `*_windows_test.go` (or any GOOS/GOARCH suffix) are
+  silently excluded from the build on other platforms — `go test` reports
+  "ok" while running nothing.

README.md

@@ -86,6 +86,7 @@ jobs:
           tz: Asia/Saigon
           include_forks: "true"
           include_private: "true"
+          include_org_repos: "false"  # "true" also counts org repos you administer (token needs read:org)
           commit_changes: "true"
 ```
 
@@ -121,7 +122,7 @@ Then embed the cards in your `README.md`:
 | `tz`               | `UTC`                            | IANA tz for the productive-time card (e.g. `Asia/Saigon`)               |
 | `start_of_week`    | `sunday`                         | First day of week for heatmap rows and weekday bars (`sunday`…`saturday`) |
 | `top_repos`        | `0`                              | Optional cap on seed repos probed for commit history (`0` = unlimited)  |
-| `commits_per_repo` | `500`                            | Max commits sampled per repo (covers last-year and all-time aggregates) |
+| `commits_per_repo` | `500`                            | Max commits sampled per repo, `0` = every commit (covers last-year and all-time aggregates) |
 | `include_forks`    | `true`                           | Include forked repos in stats and commit probing                        |
 | `include_private`  | `true`                           | Include private repos (requires PAT with `repo` scope; silently no-op otherwise) |
 | `include_org_repos`| `false`                          | Count org-owned repos you administer toward stars, repo count, languages, top-starred (needs `read:org`) |
@@ -152,6 +153,12 @@ export GITHUB_TOKEN=ghp_xxx
 ghstats -user tiennm99 -themes dracula,github_dark -tz Asia/Saigon -out output
 ```
 
+Add `-include-org-repos` to also count org-owned repos you administer:
+
+```sh
+ghstats -user tiennm99 -themes dracula -include-org-repos -out output
+```
+
 | Flag                | Default         | Description                                                            |
 | ------------------- | --------------- | ---------------------------------------------------------------------- |
 | `-user`             | *(required)*    | GitHub username                                                        |
@@ -161,7 +168,7 @@ ghstats -user tiennm99 -themes dracula,github_dark -tz Asia/Saigon -out output
 | `-tz`               | `Local`         | IANA timezone for productive-time cards                                |
 | `-start-of-week`    | `sunday`        | First day of week for heatmap rows and weekday bars (`sunday`…`saturday`) |
 | `-top-repos`        | `0`             | Optional cap on seed repos probed (`0` = unlimited)                    |
-| `-commits-per-repo` | `500`           | Max commits sampled per repo                                           |
+| `-commits-per-repo` | `500`           | Max commits sampled per repo, `0` = every commit                       |
 | `-include-forks`    | `true`          | Include forked repos in the stats                                      |
 | `-include-private`  | `true`          | Include private repos (requires `repo` PAT scope; silently no-op otherwise) |
 | `-include-org-repos`| `false`         | Count org-owned repos you administer toward stars, repo count, languages, top-starred |

action.yml

@@ -36,7 +36,7 @@ inputs:
     required: false
     default: "0"
   commits_per_repo:
-    description: Max commits sampled per repo (covers last-year and all-time aggregates)
+    description: Max commits sampled per repo, `0` for every commit (covers last-year and all-time aggregates)
     required: false
     default: "500"
   include_forks:

docs/system-architecture.md

@@ -49,7 +49,7 @@ FetchContributionsAllTime(ctx, profile, opts)
   │          TotalCommitsAllTime
   │
   ▼
-FetchProductive(ctx, profile, profile.SeedRepos, loc, commitsPerRepo)
+FetchProductive(ctx, profile, profile.SeedRepos, loc, commitsPerRepo)  // 0 = no cap
   │  commitHistoryQuery × (#seeds × pages)
   │  per commit: t = committedDate in loc
   │              ProductiveAllTime[t.Hour]++

internal/github/commit_cap_test.go

@@ -0,0 +1,27 @@
+package github
+
+import "testing"
+
+func TestReachedCommitCap(t *testing.T) {
+	cases := []struct {
+		name       string
+		seen       int
+		maxPerRepo int
+		want       bool
+	}{
+		{"under the cap", 100, 500, false},
+		{"at the cap", 500, 500, true},
+		{"past the cap", 700, 500, true},
+		// Zero means unlimited, so even a fresh repo keeps paginating. The
+		// old behavior stopped here and rendered empty commit-derived cards.
+		{"zero cap, nothing seen yet", 0, 0, false},
+		{"zero cap, deep into history", 100_000, 0, false},
+		{"negative cap treated as unlimited", 10, -1, false},
+	}
+	for _, tc := range cases {
+		if got := reachedCommitCap(tc.seen, tc.maxPerRepo); got != tc.want {
+			t.Errorf("%s: reachedCommitCap(%d, %d) = %v, want %v",
+				tc.name, tc.seen, tc.maxPerRepo, got, tc.want)
+		}
+	}
+}

internal/github/productive.go

@@ -30,9 +30,21 @@ type productiveGQL struct {
 // magnitude is irrelevant because the card renders percentages.
 const scaleFactor = 10_000
 
+// reachedCommitCap reports whether a repo has given up enough commits to stop
+// paginating. A cap of zero or less means no cap — keep going until the
+// history runs out — matching how -top-repos treats zero. Without the guard a
+// zero cap would stop before the first page and quietly empty every
+// commit-derived card.
+func reachedCommitCap(seen, maxPerRepo int) bool {
+	if maxPerRepo <= 0 {
+		return false
+	}
+	return seen >= maxPerRepo
+}
+
 // FetchProductive paginates the default-branch commit history (authored by
-// the target user) for each repo up to maxPerRepo commits, and fills two
-// parallel sets of aggregates on the Profile:
+// the target user) for each repo up to maxPerRepo commits (0 = all), and
+// fills two parallel sets of aggregates on the Profile:
 //
 //   - Last-year: p.Productive (24h histogram) and p.CommitsByLanguage
 //   - All-time:  p.ProductiveAllTime and p.CommitsByLanguageAllTime
@@ -62,7 +74,7 @@ func (c *Client) FetchProductive(ctx context.Context, p *Profile, repos []RepoIn
 		var cursor *string
 		seen := 0
 		for {
-			if seen >= maxPerRepo {
+			if reachedCommitCap(seen, maxPerRepo) {
 				break
 			}
 			owner := repo.Owner

main.go

@@ -24,7 +24,7 @@ func main() {
 		themesFlag     = flag.String("themes", "dracula", "comma-separated theme ids, or 'all'")
 		tzName         = flag.String("tz", "Local", "timezone for productive-time card (IANA name, e.g. Asia/Saigon)")
 		topRepos       = flag.Int("top-repos", 0, "optional cap on seed repos probed for commit history (0 = unlimited)")
-		perRepo        = flag.Int("commits-per-repo", 500, "max commits sampled per repo (covers both last-year and all-time aggregates)")
+		perRepo        = flag.Int("commits-per-repo", 500, "max commits sampled per repo, 0 = every commit (covers both last-year and all-time aggregates)")
 		includeForks   = flag.Bool("include-forks", true, "include forked repos in stats and commit probing")
 		includePrivate = flag.Bool("include-private", true, "include private repos (requires PAT with repo scope; silently no-op otherwise)")
 		includeOrgs    = flag.Bool("include-org-repos", false, "count org-owned repos you administer toward stars, repo count, repos-per-language and top-starred")