Created
August 30, 2026 12:08
-
-
Save mrexodia/506914176e2fb7a154860646d6a8efa7 to your computer and use it in GitHub Desktop.
Standalone reproduction for Qwen3.8 Flash Next SM121 xhigh agentic token-id-0 loop on current recipe main
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| #!/usr/bin/env python3 | |
| """Reproduce the SM121 token-id-0 (`!`) loop with a realistic agent request. | |
| WARNING: on the affected image, the xhigh request poisons/reset-crashes the TP | |
| engine. Run this only on a disposable test server; expect to restart both | |
| containers afterwards. | |
| Stdlib only. The embedded fixture is a real 31-tool coding-agent first turn, | |
| with no model weights, source checkout, or private capture required. | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import json | |
| import time | |
| import urllib.error | |
| import urllib.request | |
| FIXTURE = json.loads(r"""{ | |
| "messages": [ | |
| { | |
| "role": "system", | |
| "content": "You are an AI agent powered by DeepSeek Harness.\n\nThe DeepSeek Harness implementation checkout is at C:\\CodeBlocks\\deepseek-harness\\. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.\n\nYou are interacting with the user through the DeepSeek Harness Web GUI at http://127.0.0.1:3080. When the user refers to \"this page\", \"this GUI\", or \"this app\" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.\n\nYou are a coding agent powered by the RadixArk/Qwen3.8-Flash-Next-NVFP4 model. Your working directory is C:\\CodeBlocks\\ida-codemode-mcp\\tests.\n\nPaths prefixed with @ are files explicitly referenced by the user. Use the read tool when their contents are needed; do not claim to have inspected a file before reading it.\n\nUse the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.\n\nUse the write tool to create files or completely replace file contents. Existing files are overwritten, so read an existing file first (the default fs-observation-policy requires it) and prefer edit for targeted changes.\n\nUse the edit tool for targeted changes to existing UTF-8 text files. It replaces literal old_string with new_string; by default old_string must appear exactly once. If old_string appears multiple times, provide a more specific old_string or set replace_all to true. Read the file first (the default fs-observation-policy requires it), unless you just created or edited it in this session.\n\nUse the glob tool — not shell find — to discover files by path pattern. A pattern with no \"/\" matches basenames at any depth, so \"*\" matches every file in the tree rather than its top level. Results are files only, never directories, and include hidden and ignored files: a result that fits comes back in modification-time order, while a larger one keeps the modification-time-ordered head.\n\nUse the grep tool — not shell grep or rg — to search file contents. Use read on a matched file when you need surrounding context.\n\nNon-zero exits are reported as `[exit code: N]` markers; investigate failures before moving on. On Windows a killed process settles as `[exit code: 1]` without a signal marker; treat a bare exit 1 after an interruption as a termination, not a command failure.\n\nTrack every background job id you start. You are notified in-session when a job finishes — do not busy-poll or sleep on one; keep working on independent steps and do not duplicate a running job's work. Before giving a final answer, collect every still-relevant job with job_output (set wait: true only when you are genuinely blocked on it), and job_kill jobs that stopped mattering.\n\nUse the web_search tool to discover current information on the web. The required queries array accepts 1–4 non-empty search queries; use a one-item array for a single search. It returns an optional answer plus a list of source URLs. Use the returned source snippets when available, and cite the relevant URLs as markdown links.\n\nUse goal tools for one long-running completion objective in the current session. create_goal may infer goal intent from a direct human request in any language; do not create a goal for routine single-turn work. Call get_goal before update_goal and copy its exact goal_id and revision. After session resume or fork, an active goal is disarmed: when a human asks to continue or resume in any wording or language, use update_goal action resume to rearm it. Mark complete only when the objective is actually achieved. Mark blocked only after the same blocking condition persists for at least 3 consecutive goal rounds, and report that concrete condition in blocked_reason; difficulty, uncertainty, or useful remaining work is not blocked.\n\nUse the workflow tool ONLY when the user explicitly asks for a workflow or for large multi-agent orchestration: you write a JavaScript script (the tool description documents the exact format) that fans work out across many subagents with phases and structured results. For one or two delegations, prefer plain subagent calls.\n\nUse the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.\n\nUse subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.\n\nUse subagent_fork in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.\n\nWhen you successfully create or modify files, mention the primary outputs in your final response. To make those and any other changed-file references clickable in Web, format them as Markdown inline code using the exact file-tool path, or a basename when unique among the files changed in that turn." | |
| }, | |
| { | |
| "role": "user", | |
| "content": "Use the read tool to inspect README.md, then summarize its deployment architecture and identify two operational risks. Do not answer before using the tool." | |
| }, | |
| { | |
| "role": "user", | |
| "content": "Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\nCurrent DSH file policy: workspace-write. Any available operation enforced by the DSH file sandbox may modify files under the session workspace: \"C:\\\\CodeBlocks\\\\ida-codemode-mcp\\\\tests\". Some platform temporary areas may also be writable.\n\nApproval policy: ask. Operations that require approval may ask through the configured answerers; without an available answerer, the request fails closed." | |
| } | |
| ], | |
| "tools": [ | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ask_user_question", | |
| "description": "Ask the user a concise question when you need confirmation, a choice, or missing information before proceeding. Send one or more questions, each with a stable id that will be echoed in the answer.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "questions": { | |
| "type": "array", | |
| "description": "Questions to ask the user before continuing.", | |
| "items": { | |
| "type": "object", | |
| "additionalProperties": true, | |
| "properties": { | |
| "id": { | |
| "type": "string", | |
| "description": "Stable id for this question; echoed in the answer." | |
| }, | |
| "question": { | |
| "type": "string", | |
| "description": "The specific question to ask the user." | |
| }, | |
| "header": { | |
| "type": "string", | |
| "description": "Optional short heading for the question, such as \"Confirm\" or \"Choose Mode\"." | |
| }, | |
| "options": { | |
| "type": "array", | |
| "description": "Optional choices to show the user. If you recommend one, put it first and append \"(Recommended)\" to that label.", | |
| "items": { | |
| "type": "object", | |
| "additionalProperties": true, | |
| "properties": { | |
| "label": { | |
| "type": "string", | |
| "description": "Short user-facing option label." | |
| }, | |
| "description": { | |
| "type": "string", | |
| "description": "One sentence explaining the tradeoff or impact." | |
| } | |
| }, | |
| "required": [ | |
| "label" | |
| ] | |
| } | |
| }, | |
| "multi_select": { | |
| "type": "boolean", | |
| "description": "Whether the user may select more than one option. Defaults to false." | |
| } | |
| }, | |
| "required": [ | |
| "id", | |
| "question" | |
| ] | |
| } | |
| } | |
| }, | |
| "required": [ | |
| "questions" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "create_goal", | |
| "description": "Create one persisted same-session completion goal when the current direct human request is a long-running objective that should continue across autonomous goal rounds. You may infer that intent without requiring the user to say \"create a goal\". Do not use this for trivial single-turn work. Execution rejects non-human and subagent authority.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "objective": { | |
| "type": "string", | |
| "description": "The concrete completion objective inferred from the direct human request." | |
| }, | |
| "max_goal_rounds": { | |
| "type": "number", | |
| "description": "Optional positive safe-integer limit on automatic continuation rounds." | |
| } | |
| }, | |
| "required": [ | |
| "objective" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "edit", | |
| "description": "Edit an existing UTF-8 text file by replacing literal text.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "file_path": { | |
| "type": "string", | |
| "description": "Path to edit, resolved by the filesystem backend." | |
| }, | |
| "old_string": { | |
| "type": "string", | |
| "description": "Literal text to replace. Must match exactly." | |
| }, | |
| "new_string": { | |
| "type": "string", | |
| "description": "Literal replacement text. Use an empty string to delete the match." | |
| }, | |
| "replace_all": { | |
| "type": "boolean", | |
| "description": "Replace all matches. Defaults to false; when false, old_string must appear exactly once." | |
| }, | |
| "sandbox_permissions": { | |
| "type": "string", | |
| "description": "The wider sandbox mode this file operation needs. Only valid as a one-shot retry of an operation the sandbox just denied; requires justification and user approval.", | |
| "enum": [ | |
| "workspace-write", | |
| "danger-full-access" | |
| ] | |
| }, | |
| "justification": { | |
| "type": "string", | |
| "description": "Required with sandbox_permissions: one sentence for the user explaining why this exact file operation needs the wider access." | |
| } | |
| }, | |
| "required": [ | |
| "file_path", | |
| "old_string", | |
| "new_string" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "exit_plan_mode", | |
| "description": "Use only in plan mode. Present your plan for the user's review and, on approval, leave plan mode. Send the COMPLETE plan as markdown, starting with a # heading that names it. The user may approve (carry out the plan from your next step) or keep planning — their feedback comes back in the tool result; revise and present again.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "plan": { | |
| "type": "string", | |
| "description": "The complete plan, as markdown, starting with a # heading that names it." | |
| } | |
| }, | |
| "required": [ | |
| "plan" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "get_goal", | |
| "description": "Read the current same-session goal, including its exact id/revision, objective, phase, completed continuation rounds, round limit, blocker reason when present, and whether another continuation is armed. Call this before updating a goal.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": {} | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "glob", | |
| "description": "Find files whose paths match a glob pattern. Returns matching file paths — never directories — including hidden and ignored files (VCS metadata directories are excluded). Up to 100 paths come back in modification-time order; a larger result returns the first 100 paths in modification-time order, says so, and reports where the complete sorted list was saved. This tool does not enumerate directory entries.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "pattern": { | |
| "type": "string", | |
| "description": "Glob pattern to match file paths against (e.g. \"**/*.ts\", \"src/**/*.test.js\"). A pattern with no \"/\" matches the basename at any depth, so \"*\" and \"*.ts\" both search the whole tree; include a separator to anchor the depth." | |
| }, | |
| "path": { | |
| "type": "string", | |
| "description": "Directory to search in. Defaults to the session workspace; a relative path resolves against it." | |
| } | |
| }, | |
| "required": [ | |
| "pattern" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "grep", | |
| "description": "Search file contents with a ripgrep regular expression. Returns matching lines with line numbers, grouped by file. Returns the first 250 matches inline; a capped result reports where the complete match list was saved. Use read on a matched file for surrounding context.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "pattern": { | |
| "type": "string", | |
| "description": "Regular expression to search for (ripgrep syntax)." | |
| }, | |
| "path": { | |
| "type": "string", | |
| "description": "File or directory to search. Defaults to the session workspace; a relative path resolves against it." | |
| }, | |
| "include": { | |
| "type": "string", | |
| "description": "One glob filter for which files to search (e.g. \"*.ts\", \"*.{js,jsx}\"). Not a list; negation is not supported." | |
| } | |
| }, | |
| "required": [ | |
| "pattern" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ida_close_database", | |
| "description": "Release this MCP server's handle without disrupting other clients.\n\n If this is the final lease on a managed idalib worker, orphaned execution is\n cancelled and this call waits for the IDB to finish closing. GUI databases are\n never closed here.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "instance_id": { | |
| "anyOf": [ | |
| { | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "description": "Optional database instance id. If omitted, release the current target.", | |
| "default": null | |
| } | |
| }, | |
| "required": [] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ida_execute_python", | |
| "description": "Execute Python and return its result plus captured stdout and stderr.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "code": { | |
| "type": "string", | |
| "description": "Python code that runs against an already-open database. Call reference(query) first; do not guess the API shape. `db` is the current ida-domain Database, and `ida_domain` is also imported globally. Imports, variables, and definitions persist for this agent's database lease. A single or trailing expression is returned. For function-style code, define run(db), execute(db), or main(db); it is invoked automatically when there is no trailing expression." | |
| }, | |
| "instance_id": { | |
| "anyOf": [ | |
| { | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "description": "Optional database instance id. If omitted, use the current target.", | |
| "default": null | |
| }, | |
| "timeout": { | |
| "type": "number", | |
| "description": "Python execution timeout in seconds. This does not include the separate initial autoanalysis wait.", | |
| "default": 360 | |
| } | |
| }, | |
| "required": [ | |
| "code" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ida_list_databases", | |
| "description": "Discover registered GUI and idalib databases.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": {}, | |
| "required": [] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ida_open_database", | |
| "description": "Attach to a GUI database or shared managed idalib worker.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "path": { | |
| "type": "string", | |
| "description": "Path to a local executable or IDB. A GUI instance is used when available." | |
| }, | |
| "set_current": { | |
| "type": "boolean", | |
| "description": "Whether this database should become the default target for execute_python().", | |
| "default": true | |
| } | |
| }, | |
| "required": [ | |
| "path" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ida_reference", | |
| "description": "Look up the active ida-domain API and return a plain-text IDA reference.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "query": { | |
| "type": "string", | |
| "description": "Class, method, or reverse-engineering concept to look up in the IDA reference." | |
| } | |
| }, | |
| "required": [ | |
| "query" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ida_save_database", | |
| "description": "Explicitly save an active GUI or idalib database.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "instance_id": { | |
| "anyOf": [ | |
| { | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "description": "Optional database instance id. If omitted, save the current target.", | |
| "default": null | |
| } | |
| }, | |
| "required": [] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "interrupt_agent", | |
| "description": "Request cancellation of a background agent's current turn by its agent id. The target may be your direct child or a deeper agent created under you. Only the current turn stops: messages already queued for the agent stay parked until a later send_message, agents it started keep running, and the agent itself stays available for follow-ups. This call returns as soon as the stop request is accepted, so the target may keep running briefly; interrupting an agent that already finished is an accepted no-op.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "agent_id": { | |
| "type": "string", | |
| "description": "The agent id of the running agent to interrupt." | |
| } | |
| }, | |
| "required": [ | |
| "agent_id" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "job_kill", | |
| "description": "Request cancellation of a running background job by job id. Returns immediately; the job settles as killed once its work actually stops.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "job_id": { | |
| "type": "string", | |
| "description": "Job id returned by the tool that started the background work." | |
| }, | |
| "reason": { | |
| "type": "string", | |
| "description": "Optional short reason, recorded in the log and forwarded to the job." | |
| } | |
| }, | |
| "required": [ | |
| "job_id" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "job_list", | |
| "description": "List your background jobs (running and finished) with their ids, kinds, and statuses.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": {} | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "job_output", | |
| "description": "Read a background job. Stream jobs return only output since the previous read; final-output jobs return their result after settlement. Every response ends with `[status: ...]`. Reads are non-blocking unless `wait: true`, which waits up to the configured cap.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "job_id": { | |
| "type": "string", | |
| "description": "Job id returned by the tool that started the background work." | |
| }, | |
| "wait": { | |
| "type": "boolean", | |
| "description": "Block until the job reaches a terminal status or the timeout expires. A timed-out wait returns [status: running] and leaves the job alive." | |
| }, | |
| "timeout_ms": { | |
| "type": "number", | |
| "description": "Max wait in milliseconds (only meaningful with wait: true). Defaults to the configured wait timeout; capped by the configured maximum." | |
| } | |
| }, | |
| "required": [ | |
| "job_id" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "list_agents", | |
| "description": "List your continuable background subagents by durable id and label. Use it to recall which ones you started, not to poll for completion — you are told when one finishes. Status comes from the live registry: running means the agent is working right now, idle means it is loaded but between turns (it may be waiting on agents it started), and ready means it exists only in storage — resumable, not terminal, and not a result waiting to be collected; a `send_message` starts a new turn on the same conversation, and a direct child remains a `send_message` candidate in every status. The snapshot is not a delivery promise — `send_message` performs the authoritative check and may still fail. Children that could not be read are reported as diagnostics instead of being silently dropped. Scope `descendants` walks the whole tree below you in stable pre-order, annotating each entry with its durable direct-parent session id and depth. You may use `send_message` only for depth-1 entries; deeper entries are candidates for `interrupt_agent` only.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "scope": { | |
| "type": "string", | |
| "description": "children (default) lists direct children only; descendants walks the complete tree below you.", | |
| "enum": [ | |
| "children", | |
| "descendants" | |
| ] | |
| } | |
| } | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "pwsh", | |
| "description": "Execute a PowerShell command (`pwsh -Command`) and return its stdout/stderr. Each call runs in a fresh pwsh process: no state (cwd, variables, functions) persists between calls — pass `workdir` instead of using `cd`. Paths use native Windows form (`C:\\...`); read environment variables with `$env:NAME`. Non-zero exits are reported as `[exit code: N]`. Current harness environment facts are exposed through managed `$env:DSH_*` variables; inspect them when needed. Commands may run under a file sandbox; a blocked file operation is reported as `[sandbox: file access denied under <mode> mode]` — a policy denial, not a bug in the command; do not retry another way. Long output is truncated to its tail; the full output is saved to a file whose path is reported when available. On Windows a force-killed command settles as `[exit code: 1]` without a signal marker — treat it as an interruption, not a command failure. Set `run_in_background: true` for long-running commands: the call returns a job id immediately; read its output with `job_output` and stop it with `job_kill`. Under the Windows sandbox, read-only pwsh runs in PowerShell ConstrainedLanguage mode, while workspace-write stays in FullLanguage unless host policy says otherwise. In read-only, prefer cmdlets and core types (`[string]`, `[datetime]`, `[regex]`, `[guid]`); .NET static calls (`[System.IO.*]::`, `[math]::`), `Add-Type`, COM objects, and reflection fail with \"only core types\" errors. `-f` formatting, property access, and core cmdlets work. In both confined modes, programs cannot open named pipes, so a command that captures another program's output through piped stdio (Node.js `child_process.spawn`/`exec` with the default `stdio: 'pipe'`) fails with EPERM, while `stdio: 'inherit'` and `stdio: 'ignore'` spawns work and PowerShell's own pipelines are unaffected. That EPERM is the documented boundary: do not retry the command another way — escalate the exact command once or restructure it to avoid capturing output. Attempting a command the sandbox may deny is safe and expected: run it and read the marker rather than assuming the denial. When a command is denied and a wider mode would let it succeed, escalate immediately in the same turn — the one sanctioned exception to a denial: retry the exact same command once with `sandbox_permissions` (the narrowest wider mode that suffices) plus a one-sentence `justification`. Do not detour through chat to ask permission first — the approval prompt raised by that retry is how the user consents. If the session states approval prompts are disabled, there is no exception: a denial is final — do not set `sandbox_permissions`. Never escalate speculatively: ground the request in a real denial — normally the one this command just hit; escalating up front is fine only when this session already denied the same access. A rejected escalation is final for that command — stop and explain, never work around it — but it does not forbid attempting or escalating other commands later.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "command": { | |
| "type": "string", | |
| "description": "The PowerShell command to execute." | |
| }, | |
| "description": { | |
| "type": "string", | |
| "description": "Clear, concise description of what this command does in active voice, 5-10 words (shown in the UI). Examples: \"ls\" → \"List files in current directory\"; \"git status\" → \"Show working tree status\"; \"Get-Process\" → \"List running processes\"." | |
| }, | |
| "timeoutMs": { | |
| "type": "number", | |
| "description": "Timeout in milliseconds. The executor applies its configured default and cap, and kills the command on expiry." | |
| }, | |
| "workdir": { | |
| "type": "string", | |
| "description": "Working directory for this command. Defaults to the session workspace; a relative path is resolved against it." | |
| }, | |
| "run_in_background": { | |
| "type": "boolean", | |
| "description": "Run in the background and return a job id immediately (collect with job_output, stop with job_kill). No timeout applies." | |
| }, | |
| "sandbox_permissions": { | |
| "type": "string", | |
| "description": "The wider sandbox mode this command needs. Only valid as a one-shot retry of a command the sandbox just denied; requires justification and user approval.", | |
| "enum": [ | |
| "workspace-write", | |
| "danger-full-access" | |
| ] | |
| }, | |
| "justification": { | |
| "type": "string", | |
| "description": "Required with sandbox_permissions: one sentence for the user explaining why this exact command needs the wider access." | |
| } | |
| }, | |
| "required": [ | |
| "command", | |
| "description" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "ralph", | |
| "description": "Run a foreground fresh-agent Ralph loop toward one immutable objective. Use only when the direct human explicitly asks for Ralph or fresh-agent iteration. Each round opens a new child with no parent conversation or prior child session; the shared workspace is long-term memory, and only a bounded structured report crosses rounds. The call returns when a worker reports completion or a concrete blocker, or at the round limit. Ordinary long-running same-session work belongs to goal tools.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "objective": { | |
| "type": "string", | |
| "description": "The immutable completion objective for every fresh Ralph round." | |
| }, | |
| "maxRounds": { | |
| "type": "number", | |
| "description": "Optional positive safe-integer round cap, bounded by the deployment ceiling." | |
| } | |
| }, | |
| "required": [ | |
| "objective" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "read", | |
| "description": "Read a UTF-8 text file and return line-numbered content.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "file_path": { | |
| "type": "string", | |
| "description": "Path to read, resolved by the filesystem backend." | |
| }, | |
| "offset": { | |
| "type": "number", | |
| "description": "1-based first line to return. Defaults to 1." | |
| }, | |
| "limit": { | |
| "type": "number", | |
| "description": "Maximum number of lines to return. Defaults to 2000." | |
| } | |
| }, | |
| "required": [ | |
| "file_path" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "read_image", | |
| "description": "Read a PNG/JPEG/WebP/GIF file and return the image itself. Harness validates and downscales large supported images before the next model request, so use this tool directly instead of installing image libraries or creating thumbnails merely to inspect an image. Independent files may be read concurrently in small batches. Requires the current model to accept image input.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "file_path": { | |
| "type": "string", | |
| "description": "Path to the image file, resolved by the filesystem backend." | |
| } | |
| }, | |
| "required": [ | |
| "file_path" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "send_message", | |
| "description": "Send a message to a background subagent by its subagent id, continuing the same conversation. It becomes the subagent's next turn: if it is still working, the message waits until its current turn finishes, so it cannot redirect work already underway. This call returns no answer from the subagent — only confirmation that the message was delivered — so use it to give it more work. A failure means the message was NOT delivered.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "subagent_id": { | |
| "type": "string", | |
| "description": "The subagent id returned when the background subagent was started." | |
| }, | |
| "message": { | |
| "type": "string", | |
| "description": "The message to deliver to the subagent." | |
| } | |
| }, | |
| "required": [ | |
| "subagent_id", | |
| "message" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "skill", | |
| "description": "Load the full instructions for an available skill. Call this with the exact skill name from the session skill catalog before acting on a task that names or clearly matches that skill.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "name": { | |
| "type": "string", | |
| "description": "The exact skill name from the available skills list." | |
| } | |
| }, | |
| "required": [ | |
| "name" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "subagent", | |
| "description": "Delegate a self-contained task to a subagent (a separate agent that works in its own context) to offload focused, independent work — research, a scoped implementation, an analysis — so it does not consume this conversation's context. The subagent returns its result, not its intermediate steps. Give it a complete, standalone prompt: it does not see this conversation. This tool runs in the background by default, immediately returns a durable subagent id, and keeps the child conversation available for later turns. When that run settles, the runtime sends the parent a notice containing its outcome and any final assistant message; `send_message` starts a later turn in the same child conversation. Set `run_in_background: false` only when your next action depends on receiving the result.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "description": { | |
| "type": "string", | |
| "description": "A short (3-5 word) description of the delegated task, for display." | |
| }, | |
| "prompt": { | |
| "type": "string", | |
| "description": "The complete, self-contained task for the subagent. It does not share this conversation's context, so include everything it needs." | |
| }, | |
| "run_in_background": { | |
| "type": "boolean", | |
| "description": "Whether to run in the background and return a durable subagent id immediately. Defaults to true. Set false to wait for the result when your next action depends on it." | |
| } | |
| }, | |
| "required": [ | |
| "description", | |
| "prompt" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "subagent_fork", | |
| "description": "Delegate a task to a subagent that inherits this conversation: a child agent seeded with all completed turns so far (it does not see the current in-flight turn). Use this when the subtask builds on this conversation's context — a follow-up analysis, a review, a continuation — without consuming this conversation's context for the work itself. You receive its result, not its intermediate steps. This tool runs in the background by default, immediately returns a durable subagent id, and keeps the child conversation available for later turns. When that run settles, the runtime sends the parent a notice containing its outcome and any final assistant message; `send_message` starts a later turn in the same child conversation. Set `run_in_background: false` only when your next action depends on receiving the result.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "description": { | |
| "type": "string", | |
| "description": "A short (3-5 word) description of the delegated task, for display." | |
| }, | |
| "prompt": { | |
| "type": "string", | |
| "description": "The task for the subagent. It already sees this conversation's completed turns, so build on them freely and state only what is new." | |
| }, | |
| "run_in_background": { | |
| "type": "boolean", | |
| "description": "Whether to run in the background and return a durable subagent id immediately. Defaults to true. Set false to wait for the result when your next action depends on it." | |
| } | |
| }, | |
| "required": [ | |
| "description", | |
| "prompt" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "todo_write", | |
| "description": "Record and update a structured task list for the current work. Send the ENTIRE list every call — it REPLACES the previous list (there are no partial updates, no per-item edits). Use it to plan multi-step work and show progress: add one todo per concrete step before you start. Mark every todo being actively worked on `in_progress` — several at once when work genuinely runs in parallel (e.g. concurrent subagents or background commands), one for sequential work; while work remains, at least one task should be `in_progress`. Mark a todo `completed` the moment it is done (do not batch completions), and allow no `in_progress` item only once all work is complete. Skip the list for trivial single-step tasks. Statuses: `pending` (not started), `in_progress` (being worked on now), `completed` (finished).", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "todos": { | |
| "type": "array", | |
| "description": "The COMPLETE task list, replacing any previous list.", | |
| "items": { | |
| "type": "object", | |
| "additionalProperties": false, | |
| "properties": { | |
| "content": { | |
| "type": "string", | |
| "description": "What the task is — a short imperative line." | |
| }, | |
| "status": { | |
| "type": "string", | |
| "description": "pending (not started) | in_progress (now) | completed (done).", | |
| "enum": [ | |
| "pending", | |
| "in_progress", | |
| "completed" | |
| ] | |
| } | |
| }, | |
| "required": [ | |
| "content", | |
| "status" | |
| ] | |
| } | |
| } | |
| }, | |
| "required": [ | |
| "todos" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "update_goal", | |
| "description": "Update the exact current goal revision. edit, pause, and resume require a direct top-level human request. During an automatic continuation of the current goal, complete and blocked are also allowed. blocked is rejected before the configured minimum round count; the model remains responsible for judging that the same condition persisted across those rounds and must explain it in blocked_reason.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "goal_id": { | |
| "type": "string", | |
| "description": "Exact id returned by get_goal." | |
| }, | |
| "revision": { | |
| "type": "number", | |
| "description": "Exact positive revision returned by get_goal." | |
| }, | |
| "action": { | |
| "type": "string", | |
| "description": "edit | pause | resume | complete | blocked", | |
| "enum": [ | |
| "edit", | |
| "pause", | |
| "resume", | |
| "complete", | |
| "blocked" | |
| ] | |
| }, | |
| "objective": { | |
| "type": "string", | |
| "description": "Replacement objective; valid only with action edit." | |
| }, | |
| "max_goal_rounds": { | |
| "type": "number", | |
| "description": "Replacement cap; valid only with action edit." | |
| }, | |
| "blocked_reason": { | |
| "type": "string", | |
| "description": "Concrete blocking condition; required only with action blocked." | |
| } | |
| }, | |
| "required": [ | |
| "goal_id", | |
| "revision", | |
| "action" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "web_search", | |
| "description": "Search the web for current information. Provide 1–4 queries in the required queries array. Returns an optional summary answer and a list of source URLs.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "queries": { | |
| "type": "array", | |
| "description": "Required search queries; accepts 1–4 items and merges their results.", | |
| "items": { | |
| "type": "string" | |
| } | |
| } | |
| }, | |
| "required": [ | |
| "queries" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "workflow", | |
| "description": "Run a JavaScript workflow script that orchestrates subagents at scale. Use this for work that fans out across many independent pieces — an audit over many files, a migration, multi-angle research, adversarial verification of findings — where you write the orchestration as a script instead of delegating turn by turn.\n\nThe workflow's identity rides the `meta` parameter as JSON: required `name` (short kebab-case) and `description` strings, optional `whenToUse` string and `phases` array (`{title, detail?, provider?, model?}`). The `script` parameter is the plain JavaScript body ONLY (NOT TypeScript, and NO `export const meta` statement — meta is a parameter, not code), running with top-level await; end with `return <value>` — the value must be JSON-serializable and is this tool's result.\n\nScript-body hooks:\n- `agent(prompt, opts?): Promise<any>` — run one subagent to completion. Without `opts.schema` it resolves to the child's final text; with `opts.schema` (an object-rooted JSON Schema using ONLY type/properties/required/additionalProperties/items/enum/const/oneOf — no pattern/format/numeric bounds) it resolves to the validated object. Resolves `null` when the child fails (filter with `.filter(Boolean)`). Other opts: `label` (display), `phase` (progress group), and independent `provider`/`model` LLM target overrides (either may be provided alone). Anything else (`effort`/`isolation`/`agentType`) is rejected loudly.\n- `pipeline(items, ...stages): Promise<any[]>` — run each item through the stages independently with NO barrier between stages (prefer this for multi-stage work). Each stage receives `(prev, item, index)`. An ordinary stage throw drops that ITEM to `null` and skips its remaining stages.\n- `parallel(thunks): Promise<any[]>` — run zero-argument functions concurrently and await ALL of them (a barrier; use only when a stage genuinely needs every prior result together). A throwing thunk resolves to `null`.\n- `phase(title)` — start a progress phase; `log(message)` — narrate progress; `args` — the tool call's `args` input, verbatim.\n\nMisused hooks (bad arguments, unknown options, unsupported schemas, tripped caps) throw errors that ALWAYS kill the script — they never dissolve into a per-item `null`.\n\nConstraints: concurrency and total-agent caps apply; no filesystem, network, timers, or Node.js APIs are provided — the agents do the work, the script only coordinates them. The run executes in the foreground: this call returns when the whole script finishes.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "script": { | |
| "type": "string", | |
| "description": "The plain-JS workflow script body (top-level await allowed; NO `export const meta` statement; end with `return <json-value>`)." | |
| }, | |
| "meta": { | |
| "type": "object", | |
| "description": "The workflow identity block (plain JSON — never code).", | |
| "additionalProperties": true, | |
| "properties": { | |
| "name": { | |
| "type": "string", | |
| "description": "Short kebab-case workflow name." | |
| }, | |
| "description": { | |
| "type": "string", | |
| "description": "One-line description of what the workflow does." | |
| }, | |
| "whenToUse": { | |
| "type": "string", | |
| "description": "Optional guidance on when this workflow applies." | |
| }, | |
| "phases": { | |
| "type": "array", | |
| "description": "Optional phase declarations matched by phase() calls.", | |
| "items": { | |
| "type": "object", | |
| "additionalProperties": true, | |
| "properties": { | |
| "title": { | |
| "type": "string", | |
| "description": "The phase title phase() calls match by exact string." | |
| }, | |
| "detail": { | |
| "type": "string", | |
| "description": "Optional one-line description of the phase." | |
| }, | |
| "provider": { | |
| "type": "string", | |
| "description": "Optional provider override this phase is expected to use." | |
| }, | |
| "model": { | |
| "type": "string", | |
| "description": "Optional model override this phase is expected to use." | |
| } | |
| }, | |
| "required": [ | |
| "title" | |
| ] | |
| } | |
| } | |
| }, | |
| "required": [ | |
| "name", | |
| "description" | |
| ] | |
| }, | |
| "args": { | |
| "type": "object", | |
| "description": "Optional JSON input exposed to the script as the `args` global (wrap a bare list as a field, e.g. {\"files\": [...]}).", | |
| "additionalProperties": true | |
| } | |
| }, | |
| "required": [ | |
| "script", | |
| "meta" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| }, | |
| { | |
| "type": "function", | |
| "function": { | |
| "name": "write", | |
| "description": "Create or fully replace a UTF-8 text file.", | |
| "parameters": { | |
| "type": "object", | |
| "properties": { | |
| "file_path": { | |
| "type": "string", | |
| "description": "Path to write, resolved by the filesystem backend." | |
| }, | |
| "content": { | |
| "type": "string", | |
| "description": "Full UTF-8 text content to write." | |
| }, | |
| "sandbox_permissions": { | |
| "type": "string", | |
| "description": "The wider sandbox mode this file operation needs. Only valid as a one-shot retry of an operation the sandbox just denied; requires justification and user approval.", | |
| "enum": [ | |
| "workspace-write", | |
| "danger-full-access" | |
| ] | |
| }, | |
| "justification": { | |
| "type": "string", | |
| "description": "Required with sandbox_permissions: one sentence for the user explaining why this exact file operation needs the wider access." | |
| } | |
| }, | |
| "required": [ | |
| "file_path", | |
| "content" | |
| ] | |
| }, | |
| "strict": false | |
| } | |
| } | |
| ] | |
| }""") | |
| def history(records: int) -> str: | |
| return "".join( | |
| f"Record {i:06d}: the coding agent inspected module_{i % 97}.py, " | |
| f"ran test group {i % 89}, and observed status {i % 101} in workspace " | |
| f"shard {i % 13}; no action remains from this historical turn. " | |
| for i in range(records) | |
| ) | |
| def request(base: str, model: str, effort: str, timeout: int, records: int) -> dict: | |
| messages = [dict(message) for message in FIXTURE["messages"]] | |
| messages[1]["content"] = history(records) + ( | |
| "\nCurrent task: Use the read tool to inspect README.md, then summarize " | |
| "its deployment architecture and identify two operational risks. " | |
| "Do not answer before using the tool." | |
| ) | |
| body = { | |
| "model": model, | |
| "messages": messages, | |
| "tools": FIXTURE["tools"], | |
| "temperature": 0, | |
| "top_p": 1, | |
| "max_tokens": 512, | |
| "stream": True, | |
| "stream_options": {"include_usage": True}, | |
| # Explicitly select the template's thinking depth. `medium` is the | |
| # control; `xhigh` is the real agent setting that exposes corruption. | |
| "chat_template_kwargs": {"reasoning_effort": effort}, | |
| } | |
| req = urllib.request.Request( | |
| base + "/v1/chat/completions", | |
| data=json.dumps(body).encode(), | |
| headers={"Content-Type": "application/json"}, | |
| ) | |
| reasoning, content, tool_text, tool_names = [], [], [], [] | |
| usage, finish = None, None | |
| started = time.monotonic() | |
| try: | |
| with urllib.request.urlopen(req, timeout=timeout) as response: | |
| for line in response: | |
| if not line.startswith(b"data:"): | |
| continue | |
| raw = line[5:].strip() | |
| if not raw or raw == b"[DONE]": | |
| continue | |
| event = json.loads(raw) | |
| usage = event.get("usage") or usage | |
| for choice in event.get("choices") or []: | |
| delta = choice.get("delta") or {} | |
| reasoning.append(delta.get("reasoning_content") or "") | |
| content.append(delta.get("content") or "") | |
| finish = choice.get("finish_reason") or finish | |
| for call in delta.get("tool_calls") or []: | |
| fn = call.get("function") or {} | |
| name = fn.get("name") or "" | |
| args = fn.get("arguments") or "" | |
| if name: | |
| tool_names.append(name) | |
| tool_text.append(name + args) | |
| except urllib.error.HTTPError as exc: | |
| return { | |
| "effort": effort, | |
| "http_error": exc.code, | |
| "error_body": exc.read().decode("utf-8", "replace")[:500], | |
| "wall_s": round(time.monotonic() - started, 3), | |
| } | |
| except Exception as exc: | |
| return { | |
| "effort": effort, | |
| "transport_error": repr(exc), | |
| "wall_s": round(time.monotonic() - started, 3), | |
| } | |
| text = "".join(reasoning + content + tool_text) | |
| return { | |
| "effort": effort, | |
| "finish_reason": finish, | |
| "prompt_tokens": (usage or {}).get("prompt_tokens"), | |
| "completion_tokens": (usage or {}).get("completion_tokens"), | |
| "tool_names": sorted(set(tool_names)), | |
| "bang_count": text.count("!"), | |
| "max_bang_run": max((len(x) for x in text.split() if x and set(x) == {"!"}), default=0), | |
| "contains_10_bangs": "!!!!!!!!!!" in text, | |
| "output_head": text[:240].replace("\n", " "), | |
| "wall_s": round(time.monotonic() - started, 3), | |
| } | |
| def flush(base: str) -> None: | |
| try: | |
| urllib.request.urlopen( | |
| urllib.request.Request(base + "/flush_cache", data=b"", method="POST"), | |
| timeout=60, | |
| ).read() | |
| time.sleep(0.5) | |
| except Exception as exc: | |
| print("cache flush warning:", repr(exc)) | |
| def health(base: str) -> str: | |
| try: | |
| with urllib.request.urlopen(base + "/v1/models", timeout=10) as r: | |
| return f"HTTP {r.status}" | |
| except Exception as exc: | |
| return repr(exc) | |
| def tiny_followup(base: str, model: str) -> str: | |
| body = { | |
| "model": model, | |
| "messages": [{"role": "user", "content": "Reply with OK."}], | |
| "temperature": 0, | |
| "max_tokens": 16, | |
| } | |
| req = urllib.request.Request( | |
| base + "/v1/chat/completions", | |
| data=json.dumps(body).encode(), | |
| headers={"Content-Type": "application/json"}, | |
| ) | |
| try: | |
| with urllib.request.urlopen(req, timeout=120) as response: | |
| return f"HTTP {response.status}: {response.read().decode('utf-8', 'replace')[:200]}" | |
| except Exception as exc: | |
| return repr(exc) | |
| def main() -> int: | |
| ap = argparse.ArgumentParser() | |
| ap.add_argument("--base-url", default="http://127.0.0.1:30000") | |
| ap.add_argument("--model", default="RadixArk/Qwen3.8-Flash-Next-NVFP4") | |
| ap.add_argument("--timeout", type=int, default=900) | |
| ap.add_argument( | |
| "--records", type=int, default=4300, | |
| help="synthetic agent-history records; 4300 renders about 207k tokens", | |
| ) | |
| ap.add_argument("--xhigh-attempts", type=int, default=3) | |
| args = ap.parse_args() | |
| base = args.base_url.rstrip("/") | |
| print("WARNING: a successful reproduction can require a container restart.") | |
| print("server before:", health(base)) | |
| # A/B control: identical 31-tool request, differing only in effort. | |
| flush(base) | |
| medium = request(base, args.model, "medium", args.timeout, args.records) | |
| print("medium control:", json.dumps(medium, indent=2)) | |
| reproduced = False | |
| for attempt in range(1, args.xhigh_attempts + 1): | |
| flush(base) | |
| xhigh = request(base, args.model, "xhigh", args.timeout, args.records) | |
| print(f"xhigh attempt {attempt}:", json.dumps(xhigh, indent=2)) | |
| reproduced = bool(xhigh.get("contains_10_bangs")) | |
| if reproduced or "http_error" in xhigh or "transport_error" in xhigh: | |
| break | |
| if reproduced: | |
| # The guard at 0f95001 truncates the visible bangs, but even a tiny next | |
| # request exercises the poisoned-prefix reset path and kills the TP pair. | |
| print("immediate tiny follow-up:", tiny_followup(base, args.model)) | |
| time.sleep(2) | |
| print("server after:", health(base)) | |
| print("REPRODUCED: token-id-0 loop; the guard did not repair root cause.") | |
| return 2 | |
| print("Not reproduced in this run; server after:", health(base)) | |
| return 0 | |
| if __name__ == "__main__": | |
| raise SystemExit(main()) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment