Add task tracking to the harness

Canonical task briefs and shared discovery for ideas, studies and other work; independently accepted and checked.

TaskPhase completeDraft record

Objective and authorization

Add task tracking that includes research, ideas and other meaningful outcomes, as requested by the user in the manager chat.¹ This is authorized harness work; future task execution retains its own applicable authorization and idea/research scope gates.

Ownership and bounded contract

Owner chat: manager 01a1241c-b623-7df3-a0d7-a85aa4a02811. The manager coordinates bounded implementation and independent Astra acceptance.

Owned implementation paths: harness/file-structure.md, harness/workflow.md, harness/okf-profile.md, harness/templates/task.md.tmpl, harness/model-routing.md, scripts/workspace.py, tests/test_workspace.py, .agents/skills/maintain-harness/SKILL.md, .agents/skills/coordinate-work/SKILL.md, and this brief. README/site integration is separately owned by the manager's site worker.

Keep existing IDEA/RES briefs canonical in place. Add standalone TASK briefs and shared frontmatter-only discovery with compatible legacy row fields and per-file warnings. Do not add a registry, allocator, new skill/role/chat, bulk migration, automatic rollup, inherited authorization, external services, or operational task execution. Workers do not write shared indexes/logs or Git. Work is bounded to these paths and relevant offline checks; stop on ownership collisions or consequential rule conflicts and report to the manager. No new spending or retrieval is included; no numeric resource cap was supplied.

Deliverables and done criteria

Task placement, template and policy preserve approval, review and manager authority.
Existing idea/study and standalone task briefs appear through tasks list and status.
Targeted tests cover mixed discovery, compatible fields, deterministic warnings, frontmatter-only reads, non-brief exclusion, symlink escape and CLI equivalence.
Changed skills and read-only validators pass; independent Astra acceptance and the manager's coordinated checks/checkpoint finish the task.

Authority and dependent impact

Placement and OKF authorities add standalone tasks without new required metadata. Workflow makes existing execution gates apply regardless of task entry point. The task template, workspace discovery/tests, model-routing command guidance and two skills are updated as dependents. Existing idea/research templates, role configuration, manager charter and evidence validators need no new fields or changed authorization behavior. README/site discovery is separately coordinated. Historical inputs/scopes/receipts and existing briefs are preserved. Shared indexes/log and Git remain manager-owned.

Progress and next action

Complete: task convention, template and unified discovery implemented; Astra independently accepted the frozen implementation with no findings. Manager's isolated coordinated checkpoint passed all 72 tests and harness/knowledge checks. Git records this completion with the harness changes. Future meaningful work may use the task template; creating a record does not authorize its execution. No follow-up execution is required by this task.

Checks and results

Worker checks passed: 14 workspace tests and the full 72-test suite, read-only harness validator (0 errors), knowledge validator (317 Markdown files, 0 errors), and skill-creator quick validation of both changed skills. Astra independently reran 14 workspace tests, checked harness consistency and observed five task briefs without warnings. Manager ran workspace.py check --tests on committed HEAD plus the exact task/manager paths, excluding concurrent uncommitted site work; all 72 tests passed. This verifies the task checkpoint, not the evolving site remediation. Offline validators do not establish semantic acceptance, live runtime reload, or execution authorization. Discovery reads selected flat frontmatter scalars, not arbitrary YAML or linked bodies; generated indexes are refreshed only at the manager checkpoint.

The initiating original input is preserved. This task has no parent/subtasks; related idea/research briefs remain canonical in place.

Exact user message preserved by transcription in the manager chat; not a platform export. ↩

Sources, provenance and record details

Record type: Task
Status: draft
Work status: complete
Generated: by: codex/gpt-6.1-sol at: '2026-10-10T08:10:25Z'
Recorded checks: No verification metadata recorded.

Sources

[user-input] User request for common tasks

All record metadata

type: Task
title: Add task tracking to the harness
description: Canonical task briefs and shared discovery for ideas, studies and other
  work; independently accepted and checked.
status: draft
generated:
  by: codex/gpt-6.1-sol
  at: '2026-10-10T08:10:25Z'
x_work_status: complete
sources:
- id: user-input
  resource: /docs/project-information/management/input-2026-10-10-tasks.md
  title: User request for common tasks