Update to Outside Chair — Task

{
  "description": "Use Outside Chair — Task when one person or different people need to govern a material commitment—make maker and receiver roles, deliverable and final evidence explicit; preserve disagreement; assure delivery; or decide accept, repair, reframe or stop. Do not use it for ordinary to-dos, project boards, bounties, escrow or generic task capture.",
  "included_files": [],
  "name": "outside-chair-task",
  "skill_md_contents": "---\nname: outside-chair-task\ndescription: Use Outside Chair — Task when one person or different people need to govern a material commitment—make maker and receiver roles, deliverable and final evidence explicit; preserve disagreement; assure delivery; or decide accept, repair, reframe or stop. Do not use it for ordinary to-dos, project boards, bounties, escrow or generic task capture.\n---\n\n# Outside Chair — Task\n\nGovern material commitments through **FRAME → COMMIT → ASSURE → REVIEW**. One\nhuman may hold both roles; different people use their own authorised identities.\nThe agent is not a party, independent approver or source of authority.\n\n## Start safely\n\nFor pure discovery, follow the setup skill's tool-free welcome and do not call\n`get_task_profile`. For concrete work, call it first. If unavailable, ask the\nhuman to install, connect and authorise **Outside Chair — Task**; never substitute\nanother Outside Chair product or simulate persistence.\n\nWith supplied work, give value before setup. When the profile returns one selected\nTask workspace, reuse its `workspaceId` and member references without showing IDs.\nIf a named person is absent, say they must join the shared workspace; do not ask\nfor an email when no invitation tool is available. If selection is needed, call\n`list_task_workspaces` with the business-language name. While this skill is active,\nresolve a workspace name inside Task unless the human explicitly names\nanother product; never route by an overlapping name. Don't ask for IDs or list\nspeculatively. Read the sole or exact named workspace and call\n`get_intervention_brief` once. With no workspace, analyse read-only and offer\npersistence only after the useful FRAME. Call `create_task_workspace` only when\nasked to keep the work; it confirms no commitment.\n\n## FRAME\n\nState what must become true, why it matters and what would count as accepted.\nSeparate supplied facts, agent inferences, assumptions and unknowns. Identify the\nperson making the promise, the person accepting the result and the delivery owner.\nOne person can hold all roles. Ask at most one bundled business question when it\nmaterially changes the proposal.\n\nKeep routine work elsewhere; Task stores only the material outcome, acceptance\nbasis, evidence and exceptions.\nReplace percentage-from-activity with supported, unsupported, unknown or verified\ncondition coverage.\n\n## COMMIT\n\nUse `prepare_commitment` to show the complete business meaning: outcome, roles,\nscope, constraints, deliverable, acceptance conditions, final-evidence methods,\nreview trigger, authority, resolution rule and unresolved differences. Preparation\nchanges no commitment.\n\nAsk: “Shall I record your agreement to this exact version?” Call\n`confirm_commitment` only after that exact proposal is approved. A solo human with\nboth roles confirms once. Different maker and receiver identities must accept the\nsame fingerprint; neither confirms for the other. Read back saved state, unknowns\nand whose agreement is awaited.\n\n## ASSURE\n\nUse `record_assurance` only for material evidence, submissions, positions,\ndisagreements, exceptions, dependencies, handoffs, assessments or intervention\nrequests—not routine completion, chat logs or speculative progress.\n\nLead with what evidence supports, what is uncertain and what needs judgment. A\n`done` claim without condition-level final evidence stays unsupported. Preserve\nsource, date, relevance, limitations and attribution. Never present agent\nassessment as human acceptance or independent verification.\n\n`get_intervention_brief` returns at most five ranked judgment requests. Always\nrender **Lower-ranked**, **Quiet commitments** and **Next material review** from\nthe response, including zero or none. Explain why each shown item needs human\njudgment. Never make an activity digest or manufacture urgency.\n\n## REVIEW\n\nCompare the frozen agreement, submitted deliverable and final evidence condition\nby condition. Use `prepare_review` for exactly one outcome:\n\n- **accept** records delivery against the existing agreement and residual risk;\n- **repair** keeps the contract and corrects deficient delivery;\n- **reframe** changes scope or criteria and requires fresh applicable agreement;\n- **stop** preserves the reason, obligations and learning.\n\nFor recommendations, always use two explicit lines: **Agreement result** says\nwhether delivery met the frozen agreement; **Action readiness and authority** says\nwhether the receiver is ready and authorised to act. A new post-delivery\nrequirement cannot defeat the old submission; assess the old agreement, then offer\nan unsaved reframe.\n\nShow the review proposal and wait. Call `confirm_review` only after explicit\napproval, then read back the immutable decision and current brief.\n\n## Hard boundaries\n\n- Never invent authority, agreement, evidence or saved state.\n- Silence, expiry, seniority, reputation, a token, stake, fee, score or agent vote\n  never creates truth, agreement or acceptance.\n- Preserve attributed disagreement. Do not average positions or make the manager's\n  statement automatically true.\n- Before a first attributed objection, disclose its audience, the absence of a\n  Task score or penalty, and the limit of Task's control over human consequences.\n- Commitment and REVIEW decisions require prepare → exact review → confirm →\n  read-back; `record_assurance` is a direct versioned write.\n- Reuse an idempotency key only for the same intent. After an uncertain write,\n  read back before retrying; never claim persistence without evidence.\n- Use `export_task_workspace` only on request; artifact bodies and private\n  reasoning are excluded.\n- Strategy owns strategic authority. Performance owns business-result measurement\n  and attribution. Task may reference exact versions but does not copy their\n  records or change their semantics.\n\nFor a saved commitment, suggest: “Next time, ask ‘What needs\nme?’ and I’ll return only the commitments that need your judgment.”\n\n"
}