feat(providers): add Codebuff provider

README.md

@@ -350,20 +350,23 @@ These are starting points, not verdicts. A 60% cache hit on a single experimenta
 
 **Pi / OMP** stores sessions as JSONL at `~/.pi/agent/sessions/<sanitized-cwd>/*.jsonl` (Pi) and `~/.omp/agent/sessions/<sanitized-cwd>/*.jsonl` (OMP). Each assistant message carries token usage (input, output, cacheRead, cacheWrite) plus inline `toolCall` content blocks. CodeBurn extracts token counts, normalizes tool names to the standard set (`bash` to `Bash`, `dispatch_agent` to `Agent`), and pulls bash commands from `toolCall.arguments.command` for the shell breakdown.
 
+**Codebuff** (formerly Manicode) stores per-chat history as JSON at `~/.config/manicode/projects/<project>/chats/<chatId>/chat-messages.json`. Codebuff bills in credits rather than tokens, so CodeBurn records each completed assistant message (via `msg.credits`) and approximates cost at the public pay-as-you-go rate ($0.01 / credit). When Codebuff routes a call through an upstream provider and the stashed RunState records token-level usage (`message.metadata.runState.sessionState.mainAgentState.messageHistory[*].providerOptions`), the real tokens and LiteLLM-calculated cost take precedence. Codebuff-native tool names (`read_files`, `str_replace`, `run_terminal_command`, `spawn_agents`, etc.) normalize to the canonical set (`Read`, `Edit`, `Bash`, `Agent`). The `manicode-dev` and `manicode-staging` channels are walked automatically when present. Honors `CODEBUFF_DATA_DIR` for a custom root.
+
 **Gemini CLI** stores sessions as single JSON files at `~/.gemini/tmp/<project>/chats/session-*.json`. Each session embeds real token counts (input, output, cached, thoughts) per message. Gemini reports input tokens inclusive of cached; CodeBurn subtracts cached from input before pricing to avoid double charging.
 
 **OpenClaw** stores agent sessions as JSONL at `~/.openclaw/agents/*.jsonl`. Also checks legacy paths `.clawdbot`, `.moltbot`, `.moldbot`. Token usage comes from assistant message `usage` blocks; model from `modelId` or `message.model` fields.
 
 **Roo Code / KiloCode** are Cline-family VS Code extensions. CodeBurn reads `ui_messages.json` from each task directory in VS Code's `globalStorage`, filtering `type: "say"` entries with `say: "api_req_started"` to extract token counts.
 
-CodeBurn deduplicates messages (by API message ID for Claude, by cumulative token cross-check for Codex, by conversation/timestamp for Cursor, by session ID for Gemini, by session+message ID for OpenCode, by responseId for Pi/OMP), filters by date range per entry, and classifies each turn.
+CodeBurn deduplicates messages (by API message ID for Claude, by cumulative token cross-check for Codex, by conversation/timestamp for Cursor, by session ID for Gemini, by session+message ID for OpenCode, by responseId for Pi/OMP, by chat folder + message ID for Codebuff), filters by date range per entry, and classifies each turn.
 
 ## Environment Variables
 
 | Variable | Description |
 |----------|-------------|
 | `CLAUDE_CONFIG_DIR` | Override Claude Code data directory (default: `~/.claude`) |
 | `CODEX_HOME` | Override Codex data directory (default: `~/.codex`) |
+| `CODEBUFF_DATA_DIR` | Override Codebuff data directory (default: `~/.config/manicode`) |
 | `FACTORY_DIR` | Override Droid data directory (default: `~/.factory`) |
 | `QWEN_DATA_DIR` | Override Qwen data directory (default: `~/.qwen/projects`) |
 

package.json

@@ -23,6 +23,7 @@
     "codex",
     "opencode",
     "pi",
+    "codebuff",
     "ai-coding",
     "token-usage",
     "cost-tracking",