termcode/dist/agent/planner.js

Superior terminal AI coding agent with enterprise-grade security, intelligent error recovery, performance monitoring, and plugin system - Advanced Claude Code alternative

import { systemPrompt, userPrompt } from "./prompts.js";
import { retrieve } from "../retriever/retrieve.js";
import { buildIndex } from "../retriever/indexer.js";
import { applyUnifiedDiff } from "./diff.js";
import { commitAll } from "../tools/git.js";
import { logSession } from "../util/sessionLog.js";
import { log } from "../util/logging.js";
import { getProvider } from "../providers/index.js";
import { loadConfig } from "../state/config.js";
export async function runTask(repo, task, dry = false, model, branchName, providerId) {
    // 1) Ensure index exists (very naive check)
    try {
        await buildIndex(repo);
    }
    catch (e) {
        log.warn("index build failed:", e);
    }
    // 2) Load config and get providers
    const config = await loadConfig();
    if (!config) {
        log.error("No configuration found. Please run onboarding.");
        return;
    }
    const currentProvider = providerId || config.defaultProvider;
    const provider = getProvider(currentProvider);
    const chatModel = model || config.models[currentProvider]?.chat;
    if (!chatModel) {
        log.error(`No chat model configured for provider ${currentProvider}`);
        return;
    }
    // 3) Make a query embedding and retrieve
    let chunks = [];
    try {
        // Try to get embeddings for better context retrieval
        const embedModel = config.models[currentProvider]?.embed || config.models.openai?.embed || "text-embedding-3-small";
        let embedProvider = provider;
        // Fallback to OpenAI for embeddings if current provider doesn't support them
        if (currentProvider !== "openai" && !config.models[currentProvider]?.embed && config.models.openai) {
            embedProvider = getProvider("openai");
        }
        const qEmb = await embedProvider.embed([task], { model: embedModel });
        chunks = await retrieve(repo, qEmb[0], 10);
    }
    catch (e) {
        log.warn("Embedding/retrieval failed, proceeding without context:", e);
    }
    // 4) Get memory file if present
    let memory = "";
    try {
        memory = (await import("node:fs/promises")).readFile(`${repo}/TERMCODE.md`, "utf8").then(String);
    }
    catch { }
    // 5) Ask model for a plan + diff
    const sys = systemPrompt(String(memory || ""));
    const usr = userPrompt(task, chunks);
    const messages = [
        { role: "system", content: sys },
        { role: "user", content: usr }
    ];
    const out = await provider.chat(messages, {
        model: chatModel,
        temperature: 0.2
    });
    // 5) Extract unified diff
    const m = out.match(/---BEGIN DIFF---\n([\s\S]*?)\n---END DIFF---/);
    if (!m) {
        log.error("No diff block returned by model. Output:\n", out);
        return;
    }
    const diff = m[1];
    if (dry) {
        console.log("\n===== PROPOSED DIFF (dry-run) =====\n" + diff);
        return;
    }
    // 6) Apply diff
    const { applied, rejected } = await applyUnifiedDiff(repo, diff);
    log.info("Applied:", applied);
    if (rejected.length)
        log.warn("Rejected:", rejected);
    // 7) Commit changes
    if (!dry && applied.length > 0) {
        const commit = commitAll(repo, `termcode: ${task}`);
        if (!commit.ok)
            log.warn("Commit failed:", commit.error);
        else
            log.info("Changes committed");
    }
    // 8) Log session
    if (branchName) {
        await logSession(repo, {
            timestamp: new Date().toISOString(),
            branchName,
            task,
            diff,
            applied,
            rejected,
            model: chatModel
        });
    }
    return { applied, rejected };
}