v1.3.98

v1.3.98

Highlights

• Hooks system: Pre/post hooks for 5 lifecycle events (agent start/stop, stream start/stop, tool use) with command and HTTP types. Standardized JSON payload with HMAC signature support. Config validation with detailed error reporting.
• Context window indicator: Real-time visual indicator in the TUI composer showing context window usage with color-coded thresholds (green < 50%, orange 50–80%, red ≥ 80%).
• Friendly API error messages: Raw provider errors (401, 429, 5xx, network timeouts, context overflow) are now translated into human-readable messages with actionable advice.
• Cost tracking: New /cost slash command shows session token usage breakdown and estimated cost. Sidebar displays estimated cost or plan name for subscription users.

New Features

Hooks System

Five lifecycle events with command-based and HTTP-based hooks:

Event	Trigger
on_agent_start	Before agent loop begins
on_agent_stop	After agent loop ends (success or failure)
on_stream_start	Before each streaming turn
on_stream_stop	After each streaming turn
on_tool_use	After any tool execution

Each hook receives a standardized JSON payload via stdin with session info, event type, tool name, duration, and error details. HTTP hooks can verify authenticity via X-GGCode-Signature header (HMAC-SHA256).

Configuration in ggcode.yaml:

hooks:
  on_agent_stop:
    - type: command
      command: ["notify-send", "ggcode agent finished"]
  on_tool_use:
    - type: http
      url: https://hooks.example.com/ggcode
      method: POST
      secret: "${HOOK_SECRET}"

See docs/guide/hooks.md for the complete guide.

Cost Tracking with Coding Plan Support

The /cost slash command shows session token usage with billing-type awareness:

• Per-token models: Shows estimated USD cost with input/output rates
• Subscription plans (GitHub Copilot, Z.ai Coding Plan, Kimi, Ark, Aliyun, MiniMax, Xiaomi MiMo): Shows "included in "
• Free models (e.g., GLM-4.5-Air): Shows "free tier"
• Unknown models: Honestly shows "no pricing data available" instead of fabricated prices

Coding plan detection works via endpoint name (IsCodingPlanEndpoint) and vendor type (IsSubscriptionVendor), matching the actual built-in endpoint configuration.

New Slash Commands

Command	Description
/cost	Session token usage breakdown and estimated cost
/diff	View git diff in chat (supports --cached, --stat, <file>; 200-line truncation)
/hooks	View all configured hooks and validation status

TUI Enhancements

• Context window indicator: Color-coded usage bar in the composer area
• Sidebar cost: Estimated cost (or plan name) displayed in the session usage panel
• Restructured /help: Commands organized into 5 categories (Session & History, Model & Provider, Development, Integrations, System) with all new commands and shortcuts documented

Fixes

• Hooks config validation: Catches invalid hook configurations (empty commands, missing URLs, invalid events) at startup with detailed error messages
• HTTP context leak fix: HTTP hooks now properly cancel in-flight requests when the context is cancelled
• Pricing table non-deterministic match: Map iteration order could cause gpt-4o-2024-08-06 to match gpt-4 (rate $30/M) instead of gpt-4o (rate $2.50/M). Fixed by selecting the longest matching prefix/suffix.

Internal

• Removed all hardcoded per-token prices from the pricing table (they were sourced from unreliable training data and contained outdated model names). The table now only contains billing-type information (subscription/per-token/free) verified from official pricing pages.
• Users can add custom per-token prices via PricingTable.Merge().
• ModelRate struct extended with Type (PricingType) and Plan (display name) fields.
• New IsMetered() and IsKnown() methods on ModelRate.
• 33 new tests for hooks dispatch, runner, validation, and payload serialization.
• 14 new tests for friendly error message pattern matching.
• 15 new tests for pricing table, subscription detection, and endpoint-based coding plan lookup.

Upgrade Notes

• Hooks are disabled by default. Configure them in ggcode.yaml under the hooks key.
• Cost estimates for subscription plans show the plan name instead of dollar amounts.
• If you previously relied on per-token cost estimates for direct API models (Anthropic, OpenAI, etc.), you'll now see "no pricing data available". Configure custom rates via the pricing table API if needed.