Roles and skills

Help me add a Researcher role to my existing work team for this need: Compare options and gather evidence for a decision.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: What decision are we supporting? Which sources are permitted? How current must the evidence be? Do you want a comparison, a short brief or a source list? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/researcher.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A sourced brief or comparison with tradeoffs and unknowns. Quality check: Important claims trace to evidence; missing information stays unknown. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Writer and editor role to my existing work team for this need: Turn your material into clear drafts for a defined audience.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: Who will read this? What should they understand or do? What format, tone and length do you want? Which facts and style examples may I use? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/writer-editor.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A draft with requested format and a short explanation of material changes. Quality check: Facts match supplied inputs, tone fits the audience and unsupported claims are removed. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Project planner role to my existing work team for this need: Break an outcome into manageable steps and dependencies.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: What outcome and deadline matter? Who owns each part? What constraints or dependencies are already known? Do you need a checklist, timeline or risk review? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/project-planner.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A plan with owners, dependencies, milestones and explicitly unknown dates. Quality check: Tasks support the outcome; invented commitments and deadlines are absent. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Data analyst role to my existing work team for this need: Check and summarize a small approved dataset.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: What question should the data answer? What does each field mean? Where may the data be processed? Which calculations and output format do you need? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/data-analyst.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A reproducible summary, calculation or chart with data-quality notes. Quality check: Calculations reconcile to the source and missing or ambiguous values are explained. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Quality reviewer role to my existing work team for this need: Check an artifact against its acceptance criteria.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: Which artifact and Done when criteria should I review? What evidence may I inspect? Which errors matter most? Should I report findings only or propose corrections? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/reviewer.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: Findings with evidence, importance and suggested corrections. Quality check: Each finding points to the artifact and a criterion; accepted files are preserved. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Developer role to my existing work team for this need: Implement a small change and verify the resulting behavior.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: What behavior should change? Which repository and files are authorized? What environment and dependencies already exist? What acceptance checks prove the change? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/developer.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A scoped change, verification results and instructions for running or reverting it. Quality check: The requested behavior is tested; unrelated work is preserved and unrun checks are labeled. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a UX and UI designer role to my existing work team for this need: Make a screen or workflow easier for its intended users.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: Who uses this and what are they trying to accomplish? Which screen or workflow should I inspect? What devices and existing design conventions matter? Do you want findings, a sketch or a small implementation proposal? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/ux-designer.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A prioritized design proposal tied to user tasks and observable usability checks. Quality check: Recommendations explain the user benefit and respect the approved scope and existing design. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Accessibility reviewer role to my existing work team for this need: Review a page or document for barriers to access.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: Which page or document should I review? Who is the audience and which accessibility requirements apply? What tools and source files may I inspect? Should I report findings or propose fixes? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/accessibility-reviewer.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: Evidence-based findings and scoped corrections, with manual checks still needed clearly labeled. Quality check: Each finding includes a location and reproducible check; automated checks alone are not called certification. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.

Help me add a Training designer role to my existing work team for this need: Create a short lesson or reusable training aid.

Use my current team folder. Verify this chat's actual workspace before writes. Read AGENTS.md, POLICY.md, CONTEXT-INDEX.md and relevant existing roles. If the project differs from my intended team, stop with one exact step to open it. Preserve my Lead, active task and existing files. Do not create a second team or edit account instructions.

Ask only questions not already answered in the task or my message. Start with: Who is learning and what do they already know? What should they be able to do afterward? How much time do they have? What source material and assessment are appropriate? Also clarify permitted inputs/tools and what you may change. Explain whether my existing role or a small reusable skill could do this instead. Keep the questions in a short batch and wait for the answers needed to define the work.

Once the scope is clear, I authorize creating a bounded project-local role at roles/training-designer.md if it adds useful capability. Record purpose, approved inputs, authority, permitted tools, output, acceptance checks and stopping conditions. Expected output: A lesson, activity or job aid tied to observable learning outcomes. Quality check: Activities and assessment match the outcome; content is supported by approved material. Do not promise expertise or accuracy that has not been demonstrated.

Start with the current session applying the specialist contract. Do not imply a separate worker exists because a role file or name was created. Keep the Lead as the single writer of canonical task state and accepted outputs. Reviewer findings go to a proposal path, not over the accepted artifact. If independent execution would help, propose it separately and verify platform support before claiming it.

Index the new role in CONTEXT-INDEX.md with its use trigger. Ensure AGENTS.md can load the selected role on demand while retaining the active task and Lead ownership. Read back changes and show how I invoke the role. Do not install plugins, grant connector access, add schedulers or perform external actions as part of role creation.

Offer one small non-sensitive trial with observable Done when criteria. If my current inputs already permit that trial, perform it within scope and save its proposal to a unique task artifact path without overwriting accepted work. Otherwise ask for the missing input. Report the role path, actual loading evidence, trial result or not run, and my next action. If local writes are unavailable, return labeled files or downloads and an exact saving step, without claiming they are installed.

If copying is blocked, select the text and copy it manually.