ZCode Large Task Fixture

Use this skill to create a large local task that behaves like a long-running ZCode Agent task. It writes both persistence layers:

• App task index: ~/.zcode/v2/tasks-index.sqlite
• Agent session DB: ~/.zcode/cli/db/db.sqlite

The generated task is synthetic test data. Always back up before writing and record the manifest path.

Quick Commands

From the z-code repository root:

node .agents/skills/zcode-large-task-fixture/scripts/generate-large-task-fixture.mjs \
  --workspace "$PWD" \
  --turns 3000

Verify an existing generated task:

node .agents/skills/zcode-large-task-fixture/scripts/generate-large-task-fixture.mjs \
  --verify sess_perf_large_YYYYMMDDHHMMSS_xxxxxxxx

Remove a generated task:

node .agents/skills/zcode-large-task-fixture/scripts/generate-large-task-fixture.mjs \
  --delete sess_perf_large_YYYYMMDDHHMMSS_xxxxxxxx

Workflow

1. Confirm the target workspace path. Use the repository root unless the user names another workspace.
2. Generate with --turns 3000 for a large fixture. This creates roughly 6,000 messages and 20,000 parts.
3. Read the script output and report the sessionId, manifest path, counts, and backup directory.
4. Verify with --verify <sessionId>.
5. Tell the user to restart or refresh the app if it was already running, then open the pinned task titled Performance Fixture Large Mixed Task.

Safety

• The script uses SQLite .backup before generation unless --no-backup is passed.
• Backups are stored under ~/.zcode/perf-task-backups/<timestamp>/.
• Manifests are stored under ~/.zcode/perf-task-manifests/<sessionId>.json.
• Use --delete <sessionId> for generated fixtures. It deletes from tasks-index.sqlite and cascades the agent session rows.
• Do not delete manually unless you have checked the manifest and confirmed the task id starts with sess_perf_large_.

Coverage

Generated sessions include:

• User text and file parts
• Assistant text and reasoning parts
• Tool parts with completed, running, pending, and error states
• Bash, Read, Grep, Glob, Edit, Write, TodoWrite, AskUserQuestion, Agent, WebSearch, and WebFetch tool shapes
• Patch, compaction, subtask, agent, retry, step-start, and step-finish parts
• Todo rows, session target rows, model usage rows, turn usage rows, and task index searchable text