@tylerho/pi-expanded-footer

Replaces the default footer with a Claude Code-style layout: model, context usage, cost, branch, worktree.

Install

pi install npm:@tylerho/pi-expanded-footer

Expanded Footer

Replaces pi's default statusline footer with a Claude Code-style layout: model · effort | context usage | cost on line 1, path · branch · worktree | token stats on line 2, and extension statuses on line 3. It exists so the session state Tyler actually watches — model, cache countdown, context %, cumulative token/cost, current git branch and worktree — is visible at a glance, with Claude's color scheme instead of pi's defaults.

Key concepts

• One setFooter per session. On session_start the extension captures ctx once and calls ctx.ui.setFooter(...) with a render closure. The closure reads ctx.sessionManager.getEntries() and ctx.getContextUsage() fresh on every render, so the footer reflects the live session. The dispose() returned by the closure clears the branch-change subscription and the 1-second ticker.
• Re-renders are event + timer driven. footerData.onBranchChange(() => tui.requestRender()) re-renders when the git branch changes; a 1s setInterval re-renders so the cache countdown ticks. Both are torn down in dispose().
• Usage is aggregated across the whole session. The render walks sessionManager.getEntries() and sums usage from assistant messages (input/output/cacheRead/cacheWrite/cost), toolResult messages (cost only), and branch_summary/compaction entries (all counters). Cache hit rate is computed from the latest assistant message: cacheRead / (input + cacheRead + cacheWrite).
• Context display mirrors Claude's. {tokens}tokens [{percent}%], colored success/warning/error at 34%/67% thresholds. Windows >200k render [{percent}% {window} | {pct200k}% 200k]. Falls back to ? tokens [?] when the percent is unknown; contextWindow falls back to ctx.model?.contextWindow ?? 200_000.
• Extension statuses ride the footer API. Other extensions call ctx.ui.setStatus(key, text); the footer reads them via footerData.getExtensionStatuses(). The subagents key is special-cased onto its own rail directly under line 1 (running/done counts visible at a glance); every other key is sorted alphabetically, whitespace-sanitized, and joined onto line 3. This is why any extension status shows up in the footer without this extension knowing about it.
• Module-level state survives sessions. currentThinkingLevel, cacheTimerEnabled, cacheConfig, and the ticker handle are module-scope, so they persist across /new//resume but reset on /reload. currentThinkingLevel starts at "off" and is overwritten by the thinking_level_select event; it only renders when the active model has reasoning.
• Git worktree detection shells out to git rev-parse --git-dir (2s timeout, silent failure) and matches /worktrees/<name> to show · wt:<name> on line 2.
• Width safety everywhere. Lines are built then trimmed with truncateToWidth/visibleWidth from pi-tui; line 2 keeps the right-side stats and, when the path doesn't fit, middle-truncates the path (head…tail) so the · branch and · wt:<name> suffixes stay visible. It drops the worktree suffix first, then shrinks the branch, before falling back to tail truncation.

API

Command

| Command | Purpose | |---|---| | /cache-timer | Toggles the cache countdown shown at the left of the token stats. No args. Flips module-level cacheTimerEnabled, notifies via ctx.ui.notify("Cache timer enabled|disabled", "info"). |

Events

| Event | Handler | Why it matters | |---|---|---| | thinking_level_select | Stores event.level in module-level currentThinkingLevel. | Keeps the · <effort> tag on line 1 in sync with the actual thinking level; only shown when the model has reasoning. | | session_start | Captures ctx, detects the worktree, loads cache config, installs the setFooter render closure. | The footer is per-session; without this the closure would hold a stale ctx. |

Footer data surface (read from footerData in the render closure)

• onBranchChange(cb) — subscribe to git branch changes; tui.requestRender() is called from it.
• getGitBranch() — branch name for line 2 (yellow).
• getExtensionStatuses() — map of ctx.ui.setStatus() keys → text; subagents gets its own rail, the rest form line 3.

Config file

• ~/.pi/agent/cache-timer.json (or $PI_CODING_AGENT_DIR/cache-timer.json) — optional TTL overrides read by loadConfig():

  { "defaultTtlMs": 300000, "providers": { "deepseek": 14400000, "google": 3600000 } }

  providers entries override the built-in per-provider TTL table entirely; defaultTtlMs is the fallback for unknown providers. The TTL tables themselves live in lib/cache-timer.ts (anthropic 5m/1h-long, openai 24h, gpt-5.6+ 30m, deepseek/groq 2h, google 1h, bedrock 5m/1h-long, openrouter derived from upstream slug).

Exports

• export default function (pi: ExtensionAPI) — the only export; standard extension entry point. No named exports, no tools, no shortcuts.

Examples

1. Model shows thinking effort — with a reasoning model active, line 1 renders like claude-sonnet-5 · high | 12.3k tokens [6.2%] | $0.041; effort comes from thinking_level_select events and is omitted for non-reasoning models.
2. Large-context window display — on a 1M-token window at 30%: gpt-5.6 · high | 300k tokens [30.0% 1M | 150% 200k] | $0.132 (percent of the 1M window plus the equivalent percent of 200k).
3. Cache timer inline stats — line 2 right side shows cache 4m/5m ↑1.2k ↓305 R4.3k W1.1k CH79.6% (timer → tokens → cache read/write → hit rate, all dim). With DeepSeek active the TTL renders as 2h; after the window passes it flips to cache expired; switching models shows cache invalidated.
4. Extension statuses — after ctx.ui.setStatus("prompt-stash", "📋 Draft saved") from prompt-stash and a running subagent batch, the footer is: line 1 (model/ctx), a subagents rail under it (2 running · 1 done), line 2 (path/branch/stats), line 3 (📋 Draft saved).
5. Disable the timer — /cache-timer removes the cache … token from line 2 and notifies "Cache timer disabled"; the rest of the footer is unaffected.