fix(plugin): move session-start to chat.message to fix Jinja system-message error

experimental.chat.system.transform fires for task-spawned subagent sessions
after the task prompt (user message) is already in the conversation. Pushing
to output.system at that point places a system message at a non-zero position,
which Qwen3.6's strict Jinja chat template rejects with 'System message must
be at the beginning.'

Move session-start.sh injection to chat.message (first turn guard via
initializedSessions). Injected as a synthetic text part via unshift(), which
has no position constraints.
This commit is contained in:
Brydon DeWitt 2026-05-22 23:05:16 -04:00
parent 5c1225744a
commit f0d21e9895

View File

@ -1,12 +1,12 @@
import type { Plugin, TextPart } from '@opencode-ai/plugin';
import { resolve, dirname } from 'node:path';
import { fileURLToPath } from 'node:url';
import type { Plugin, TextPart } from "@opencode-ai/plugin";
import { resolve, dirname } from "node:path";
import { fileURLToPath } from "node:url";
/**
* Agent support plugin for Remnant.
*
* Responsibilities:
* 1. experimental.chat.system.transform session-start.sh (once per session)
* 1. chat.message (first turn) session-start.sh (once per session)
* 2. chat.message user-prompt-submit.sh (each turn)
* 3. tool.execute.before pre-tool-use.sh (project policy)
* 4. tool.execute.after post-tool-use.sh + context pressure warning
@ -26,7 +26,10 @@ const PRESSURE_THRESHOLD = 0.7; // 70%
const LOCAL_WORKER_MAX_TOKENS = 1500;
const LOCAL_ORCHESTRATOR_MAX_TOKENS = 2500;
function truncate(text: string, maxTokens: number): { text: string; truncated: boolean } {
function truncate(
text: string,
maxTokens: number,
): { text: string; truncated: boolean } {
const maxChars = maxTokens * CHARS_PER_TOKEN;
if (text.length <= maxChars) return { text, truncated: false };
return {
@ -42,14 +45,16 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
// This makes the plugin work both as a project-local plugin and as a global
// plugin installed via install.sh — in either case, hooks live in ../../hooks/
// relative to this file in the .agents/frameworks/opencode/ directory.
const hooksDir = resolve(dirname(fileURLToPath(import.meta.url)), '../../hooks');
const hooksDir = resolve(
dirname(fileURLToPath(import.meta.url)),
"../../hooks",
);
// Running cumulative context size estimate (characters)
let contextCharsUsed = 0;
// Track sessions that have had session-start injected (system.transform fires every turn)
// Track sessions that have had session-start injected (fires once per session)
const initializedSessions = new Set<string>();
/** Parse the additionalContext string from a hook's JSON output. */
function parseAdditionalContext(hookOutput: string): string | undefined {
try {
@ -62,7 +67,10 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
}
}
async function runHook(scriptName: string, stdinJson?: string): Promise<string> {
async function runHook(
scriptName: string,
stdinJson?: string,
): Promise<string> {
const script = `${hooksDir}/${scriptName}`;
try {
const proc = stdinJson
@ -72,61 +80,85 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
} catch (_error) {
// DEBUG: log hook failures so silent catches don't hide enforcement bugs
try {
const fs = await import('node:fs');
const fs = await import("node:fs");
fs.appendFileSync(
'/tmp/plugin-hook-errors.log',
JSON.stringify({ ts: new Date().toISOString(), script, error: String(_error) }) + '\n'
"/tmp/plugin-hook-errors.log",
JSON.stringify({
ts: new Date().toISOString(),
script,
error: String(_error),
}) + "\n",
);
} catch (_e) {
// ignore
}
// Hooks are advisory — never block on hook failure
return '';
return "";
}
}
return {
// ── 1. Session start: inject project state into system prompt ────────────
// Uses system.transform rather than the defunct session.created event.
// Guarded by initializedSessions so it runs exactly once per session.
'experimental.chat.system.transform': async (input, output) => {
const sessionID = input.sessionID ?? 'unknown';
if (initializedSessions.has(sessionID)) return;
initializedSessions.add(sessionID);
const hookOutput = await runHook('session-start.sh');
const context = parseAdditionalContext(hookOutput);
if (context) output.system.push(context);
},
// ── 1 & 2. Session start + user prompt ──────────────────────────────────
// Session-start was previously injected via experimental.chat.system.transform
// (pushing to output.system). That caused a Jinja "System message must be at
// the beginning" error on Qwen-family local models when the orchestrator spawns
// a subagent via `task`: system.transform fires after the task prompt (a user
// message) is already in the conversation, so the system push lands at a
// non-zero position. Injecting as a synthetic text part on the first
// chat.message turn avoids the position constraint entirely.
"chat.message": async (input, output) => {
const sessionID = input.sessionID ?? "unknown";
// Session-start injection — runs exactly once per session, prepended so it
// reads before the user-prompt-submit nudges on the first turn.
if (!initializedSessions.has(sessionID)) {
initializedSessions.add(sessionID);
const startOutput = await runHook("session-start.sh");
const startContext = parseAdditionalContext(startOutput);
if (startContext) {
output.parts.unshift({
id: `prt_${crypto.randomUUID()}`,
sessionID: input.sessionID,
messageID: input.messageID ?? crypto.randomUUID(),
type: "text",
text: startContext,
synthetic: true,
});
}
}
// ── 2. User prompt: task capture + CURRENT QUESTION + nudges ──────────────
// Equivalent to Copilot's UserPromptSubmit hook.
'chat.message': async (input, output) => {
const promptText = output.parts
.filter((p): p is TextPart => p.type === 'text')
.map(p => p.text)
.join('\n');
const hookOutput = await runHook('user-prompt-submit.sh', JSON.stringify({ prompt: promptText }));
.filter((p): p is TextPart => p.type === "text")
.map((p) => p.text)
.join("\n");
const hookOutput = await runHook(
"user-prompt-submit.sh",
JSON.stringify({ prompt: promptText }),
);
const context = parseAdditionalContext(hookOutput);
if (context) {
output.parts.push({
id: `prt_${crypto.randomUUID()}`,
sessionID: input.sessionID,
messageID: input.messageID ?? crypto.randomUUID(),
type: 'text',
type: "text",
text: context,
synthetic: true,
});
}
},
// ── 3. Pre-tool-use ─────────────────────────────────────────────────────
'tool.execute.before': async (input, output) => {
"tool.execute.before": async (input, output) => {
const toolName = input.tool as string;
// ── read guards ───────────────────────────────────────────────────
if (toolName === 'read') {
const args = (output.args ?? {}) as { filePath?: string; offset?: number; limit?: number };
const filePath = args.filePath ?? '';
if (toolName === "read") {
const args = (output.args ?? {}) as {
filePath?: string;
offset?: number;
limit?: number;
};
const filePath = args.filePath ?? "";
// package.json read guard:
// Reading workspace package.json files auto-loads nested AGENTS.md files
@ -134,7 +166,7 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
// Block package.json reads under apps/ and packages/ only.
if (/(^|\/)(apps|packages)\/[^/]+\/package\.json$/.test(filePath)) {
throw new Error(
'BLOCKED: Reading workspace package.json files auto-loads nested AGENTS.md files and exhausts the 32K context. Use `grep_search` to find the specific field you need (e.g. a dependency version or script name) instead of reading the whole file.'
"BLOCKED: Reading workspace package.json files auto-loads nested AGENTS.md files and exhausts the 32K context. Use `grep_search` to find the specific field you need (e.g. a dependency version or script name) instead of reading the whole file.",
);
}
@ -146,7 +178,7 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
// Directory reads (e.g. `Read .`) never carry a limit — skip the guard.
let isDirectory = false;
try {
const { statSync } = await import('node:fs');
const { statSync } = await import("node:fs");
isDirectory = statSync(filePath).isDirectory();
} catch (_error) {
// path doesn't exist or inaccessible — treat as file
@ -158,7 +190,7 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
throw new Error(
isDocsFile
? `BLOCKED: Unbounded read (no limit) is prohibited. Specify offset and limit to read in ≤500-line chunks for docs/ files.`
: `BLOCKED: Unbounded read (no limit) is prohibited. Use grep_search first to find the relevant section, then read with offset and limit in ≤50-line chunks.`
: `BLOCKED: Unbounded read (no limit) is prohibited. Use grep_search first to find the relevant section, then read with offset and limit in ≤50-line chunks.`,
);
}
const lineLimit = isDocsFile ? 500 : 50;
@ -166,7 +198,7 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
throw new Error(
isDocsFile
? `BLOCKED: Read more than 500 lines at once is prohibited for docs/ files. Use offset and limit to paginate in ≤500-line chunks.`
: `BLOCKED: Read more than 50 lines at once is prohibited. Use offset and limit to paginate in ≤50-line chunks. For docs/ files the limit is 500 lines. Use grep_search first to find the right offset.`
: `BLOCKED: Read more than 50 lines at once is prohibited. Use offset and limit to paginate in ≤50-line chunks. For docs/ files the limit is 500 lines. Use grep_search first to find the right offset.`,
);
}
}
@ -177,12 +209,12 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
// or long inventories inline in a task prompt causes "Unterminated string"
// parse errors. Cap task prompts at 1200 chars — workers should be told
// WHICH files to read, not given the contents inline.
if (toolName === 'task') {
if (toolName === "task") {
const args = (output.args ?? {}) as { prompt?: string };
const prompt = args.prompt ?? '';
const prompt = args.prompt ?? "";
if (prompt.length > 1200) {
throw new Error(
`BLOCKED (task prompt too long: ${prompt.length} chars, max 1200): Task prompts must not embed file contents, dependency lists, or long context inline — this causes JSON parse failures. Instead, tell the worker WHICH files to read and WHAT to do. Example: "Read the root package.json and all workspace package.json files, then update the Technology Stack section in README.md to match."`
`BLOCKED (task prompt too long: ${prompt.length} chars, max 1200): Task prompts must not embed file contents, dependency lists, or long context inline — this causes JSON parse failures. Instead, tell the worker WHICH files to read and WHAT to do. Example: "Read the root package.json and all workspace package.json files, then update the Technology Stack section in README.md to match."`,
);
}
}
@ -194,31 +226,37 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
tool_name: toolName,
tool_input: output.args ?? {},
});
const hookResult = await runHook('pre-tool-use.sh', hookInput);
const hookResult = await runHook("pre-tool-use.sh", hookInput);
// If the hook emitted a deny decision, surface it as an error
if (hookResult.includes('"permissionDecision": "deny"')) {
const match = hookResult.match(/"permissionDecisionReason":\s*"([^"]+)"/);
const reason = match?.[1] ?? 'Blocked by project policy (pre-tool-use hook).';
const match = hookResult.match(
/"permissionDecisionReason":\s*"([^"]+)"/,
);
const reason =
match?.[1] ?? "Blocked by project policy (pre-tool-use hook).";
throw new Error(reason);
}
},
// ── 4. Post-tool-use ────────────────────────────────────────────────────
'tool.execute.after': async (input, output) => {
"tool.execute.after": async (input, output) => {
const response = output.response as string | undefined;
if (typeof response === 'string') {
if (typeof response === "string") {
// a) Response truncation — local agents (build/orchestrator) and any ollama/ model;
// orchestrator gets a higher limit since it only reads, not edits.
const agentName = typeof input.agent === 'string' ? input.agent : '';
const agentName = typeof input.agent === "string" ? input.agent : "";
const isLocalAgent =
agentName === 'build' ||
agentName === 'orchestrator' ||
(typeof input.model === 'string' && input.model.startsWith('ollama/'));
agentName === "build" ||
agentName === "orchestrator" ||
(typeof input.model === "string" &&
input.model.startsWith("ollama/"));
if (isLocalAgent) {
const isOrchestrator = agentName === 'orchestrator';
const maxTokens = isOrchestrator ? LOCAL_ORCHESTRATOR_MAX_TOKENS : LOCAL_WORKER_MAX_TOKENS;
const isOrchestrator = agentName === "orchestrator";
const maxTokens = isOrchestrator
? LOCAL_ORCHESTRATOR_MAX_TOKENS
: LOCAL_WORKER_MAX_TOKENS;
const { text: truncated } = truncate(response, maxTokens);
output.response = truncated;
}
@ -242,13 +280,13 @@ export const AgentSupportPlugin: Plugin = async ({ $, directory }) => {
tool_input: input.args ?? {},
tool_response: (output.response as string).slice(0, 500), // truncated for hook
});
await runHook('post-tool-use.sh', hookInput);
await runHook("post-tool-use.sh", hookInput);
}
},
// ── 5. Pre-compact: export state before context summarization ─────────────
'experimental.session.compacting': async (input, output) => {
await runHook('pre-compact.sh');
"experimental.session.compacting": async (input, output) => {
await runHook("pre-compact.sh");
output.prompt = `
You are a context summarizer for coding sessions. Summarize only the conversation history given do not answer it.