Context Info

📄 README

Why. Replaces pi’s default footer with a rich dashboard: git branch, model name, token usage with color thresholds, TPS during streaming, cache hit rate, session name, trust status, thinking level, live timer, and tool call counter.

How it works. Creates FooterState on session start, reads config from .pi/settings.json (top-level settings: quietStartup, contextStatusBar.showTps). Detects git worktree name, captures session name and trust status. Installs custom footer via ctx.ui.setFooter() (TUI only). Updates reactively on model_select, thinking_level_select, turn_end, message_end (token+cache stats), message_update (TPS sampling through deduplicated key extraction), tool_execution_end. Timer updates session duration display every second. Session_shutdown stops timer.

Registers three /explain-* commands for listing discovery:

• /explain-extensions — Lists all active extensions with descriptions
• /explain-prompts — Lists all available prompt templates
• /explain-skills — Lists all available skills

Also provides setSupervisorIssueData / clearSupervisorIssueData exported API for the supervisor pipeline to display current issue number/title/repo in the TUI footer.

Location: .pi/extensions/context-info/

Details

Architecture

Reactive footer system with event-driven updates and timer:

├── index.ts            # Entry: event hooks, state management, /explain-* commands
├── footer.ts           # installFooter: builds TUI footer component tree
├── footer-state.ts     # FooterState: mutable state container with render triggers
├── config.ts           # Load config from .pi/settings.json (quietStartup, contextStatusBar)
├── types.ts            # ThresholdEntry, TpsSample, FooterConfig interfaces
├── git-helpers.ts      # Worktree name detection from git rev-parse
├── telemetry.ts        # tryEmit: lightweight telemetry emission
├── extensions.ts       # List active extensions from .pi/extensions/
├── prompts.ts          # List available prompt templates
├── skills.ts           # List available skills
├── explain.ts          # createExplainCommand: /explain-* command factory with word-wrap
├── cheasee-pi-info.ts  # /cheasee-pi-info command with repo version display
└── test/               # Unit tests

Footer State Machine

flowchart TD
    A[session_start] --> B[FooterState.resetProperties]
    B --> C[loadConfig]
    C --> D{config null?}
    D -- yes --> E[clear UI: setFooter undefined, setStatus undefined, stopTimer]
    D -- no --> F[set worktreeName, sessionName, trustStatus]
    F --> G[installFooter: build TUI component]
    G --> H[startTimer: 1s interval]
    
    H --> I{Event received}
    I -- model_select --> J[update model name, contextWindow, cacheHitRate]
    I -- thinking_level_select --> K[update thinkingLevel]
    I -- turn_end --> L[re-render footer]
    I -- message_end --> M[update token usage, cache stats]
    I -- message_update --> N[sample TPS]
    I -- tool_execution_end --> O[increment toolCallCount]
    I -- session_shutdown --> P[stopTimer, dispose]

Footer Component Tree

The TUI footer renders a single-line status bar with adaptive segments:

[Git branch]  [Model name]  [Thinking level]  [⏱ timer]  [Tokens used/max]  [TPS]  [Cache hit rate]  [Session name]  [Trust status]  [Tool calls]

Each segment is implemented as a reactive Widget that updates when its backing Reactive value changes. The FooterState object holds all reactive values and calls installFooter() (mutating FooterConfig) which triggers ctx.ui.setFooter() re-render.

Key Design Decisions

• FooterState lifecycle — Created on session_start, disposed on session_shutdown. Previous state is disposed before creating new one to prevent stale ctx closures after reload/newSession/fork/switchSession.
• Working indicator — Custom dot pulse (· → • → ● → •) instead of standard spinner. Set in session_start for TUI mode.
• TPS sampling — On message_update event, samples streaming output tokens. Uses deduplicated key extraction to prevent double-counting. Sampled every ~200ms during streaming.
• Cache hit rate — Computed from cacheRead / (cacheRead + cacheWrite) on each message_end. Rounded to integer percentage. Reset on model change (provider/model-specific cache keys).
• Quiet startup — When quietStartup: true in config, startup hint and working indicator are suppressed.
• Explain commands — /explain-extensions, /explain-prompts, /explain-skills render inline lists. Widgets auto-cleared on first input/before_agent_start/user_bash event.
• Supervisor integration — Exported setSupervisorIssueData/clearSupervisorIssueData allow supervisor pipeline to display current issue number/title/repo in the TUI footer. Uses function reference (stateRef) set on session_start.
• Timer — startTimer() sets a setInterval at 1s. stopTimer() clears it. Timer updates FooterConfig.timer value and triggers re-render.

Event → Footer Update Map

Event	Fields Updated
session_start	worktreeName, sessionName, trustStatus, sessionId, contextWindow
model_select	modelName, contextWindow, cacheHitRate, sessionName
thinking_level_select	thinkingLevel
turn_end	sessionName (re-read)
message_end	tokensUsed, tokensMax, cacheRead, cacheWrite, cacheHitRate
message_update	TPS samples
tool_execution_end	toolCallCount
session_shutdown	Timer stopped, state disposed

Testing

Tests cover:

• Footer state: all reactive values, mutations, reset
• Event handlers: each event → correct field update
• Timer: start/stop/interval behavior
• TPS sampling: deduplication, accumulation, average computation
• Explain commands: extension/prompt/skill listing, error handling
• Config loading: all settings paths, missing file, invalid JSON
• Supervisor integration: set/clear issue data functions