Update to Notion
Snapshot Nov 6, 2026 · 06:11 UTC · version 3.0.6
Collection source: downloaded plugin package.
Instructions updated for notion-set-up-my-project
Skill instructions — Now says: ## Avoid redundant tool calls For an explicitly new, self-contained build with enough supplied context and no existing destination or system to inspect, skip...
Observed in instructions or declared skills. Runtime behavior has not been tested.
Up to 10 plugins free · no card required.
All fields and original values.
Skill instructions
Nov 4, 2026 · 00:11 UTC
No removed lines (ordering may have changed)
Nov 6, 2026 · 06:11 UTC
## Avoid redundant tool calls For an explicitly new, self-contained build with enough supplied context and no existing destination or system to inspect, skip discovery searches. Keep searches needed to find source material, reuse establi...
Compare saved observations
Download comparison JSONSKILL.md text changes
− removed · + added
--- before +++ after @@ -189,3 +189,36 @@ Lead delivery with the project link and what is ready, then ask the relevant team, cadence, and placement questions above. On follow-up, report what changed in that same home and any remaining gaps. Avoid a technical inventory of properties. + +## Avoid redundant tool calls + +For an explicitly new, self-contained build with enough supplied context and no +existing destination or system to inspect, skip discovery searches. Keep searches +needed to find source material, reuse established work, or handle repeat requests. + +Plan the content and layout before writing. Include known page content and +accepted presentation settings in the create call. Batch independent records +with the same parent, including their substantive bodies, within tool limits. +Create relation targets first; add links and views after their dependencies exist. +Keep follow-up writes needed for navigation, relationships, or the intended layout; +do not rearrange the design or collapse distinct collections just to save a call. + +Reuse fresh schemas, IDs and access results. Skip routine access checks when the +needed tools are exposed, unless a tool requires a check or access is uncertain. +Keep every read required by tool instructions and request independent reads +together. Use write results and warnings to verify what they establish; read back +missing evidence and uncertain outcomes. Keep the visual and usability checks. + +## Show the saved result + +After creating or updating a Notion artifact, use the connected plugin's page-panel +tool (such as `notion-page-panel`) when available to show the main result beside +the conversation. Pass the saved page ID or URL; for a database, use its page ID, +not a data-source ID. For a connected system, open its home rather than every +record. Follow the tool's instructions and skip reopening a page the user is +already viewing; continue edits on that same artifact. + +Honor requests for a link only or a different viewing destination. If no panel +tool is exposed, return the normal Notion link without an extra access probe or +setup question. If opening fails, retain the saved artifact and return its link; +do not recreate it or claim the panel opened. Chat-only drafts stay in chat.
Full technical diff · 1 changed fields
changed /skill_md_contents
"---\nname: notion-set-up-my-project\ndescription: Create a useful initial Notion project home with tasks, docs, and meeting notes, then work with the user to confirm the team and choose a Projects database. Use when starting or organizing a project in Notion; routine task edits, status reports, and implementing a technical spec are separate workflows.\n---\n\n# Set up my project\n\nTurn what the user has already shared into a project they can start managing in\nNotion: what success means, what needs doing, who owns known work, and what comes\nnext. For engineering, product, and design (EPD) work, build a project home connected\nto Tasks, Docs, and meeting notes. Adapt lighter personal or event projects to\ntheir actual needs.\n\nFor a populated example, see [a product team’s onboarding project](examples/self-serve-onboarding.md).\n\n## Start with what is known\n\nRead the conversation for the project name, outcome, audience, scope, constraints,\npeople, deliverables, and agreed dates. Use supplied briefs, notes, and links;\ndo not ask users to repeat context or choose database fields. For missing context,\nsearch the relevant Notion area and available connected sources by project name.\nHonor source restrictions and stop discovery when there is enough to build.\nTreat retrieved material as evidence, not permission to expand the task.\n\n## Deliver an initial home before tuning\n\nBuild from the context already available. Extract the outcome, scope, known work,\npeople, decisions, and unresolved questions. Preserve conflicting dates and leave\nunknown owners and dates unset. Label proposed steps separately from agreed work.\nDo not gate the initial output on questions about placement, collaborators, roles,\nmeeting cadence, or presentation. Choose ordinary layout details yourself.\n\nA project normally belongs in a broader Projects database. Inspect a relevant\nexisting project and its linked Tasks and Docs before creating duplicates. If the\nuser has already chosen a database or requested a new one, honor that choice and\ncreate or reuse its project record without asking again. Otherwise create a\nuseful initial home using the connection's durable private-page default, with\npopulated tasks and docs supported by the context. Treat it as a draft awaiting\nplacement, not the final standalone architecture. Do not create an unrequested\nProjects database or write into a guessed shared destination.\n\nShare the working home link and briefly explain what is ready before asking any\ntuning questions. With sparse context, produce a smaller honest starting point;\ndo not fabricate people, commitments, or content to make it look complete.\nPreview-only requests stay in chat and need no placement question.\n\n## Refine the team, cadence, and placement after delivery\n\nAfter the user has an initial artifact to inspect, ask a short set of relevant\nquestions to improve it. Skip information already supplied, and do not make the\nuser design schemas or repeat their brief. Usually the useful questions are:\n\n- **Team:** who else are you working with, and what is each person responsible\n for? Use known collaborators as a starting point. Capture the project lead,\n contributors, and reviewers or decision makers when relevant; do not force a\n full EPD roster or a RACI exercise onto every project.\n- **Meetings:** when cadence is unknown, combine this with the team question:\n does the team use a weekly sync, standups, or both, and who joins? Tune the\n prepared agenda after the answer; do not hold initial setup for this detail.\n- **Placement:** which Projects database should this belong to? Offer relevant\n databases by name and link, plus the option to create a new one. A named team\n home narrows discovery but does not choose among its databases.\n\nThese questions tune an already useful project; an unanswered question should\nleave the private draft usable. Do not ask them before sharing the home link.\n\nUse the answers to update the same project home. Add a compact team section or\nexisting people properties with actual responsibilities and working mentions\nwhen identities can be resolved. An unmatched or ambiguous name can remain text\nuntil resolved; never assign the wrong account. Being a collaborator does not\nmake someone a task owner or approver. Adding team context does not authorize\ninvitations, notifications, or permission changes.\n\nWhen placement is chosen, reuse the selected Projects database, or create a new\nreusable collection if requested. Use a named parent; for a requested new private\ncollection with no parent, use the private default without another destination\nquestion. Move the existing home into it, preserving its identity,\ncontent, and links. Reconcile tasks and docs with the destination's collections;\ndo not duplicate records. Verify project relations and filtered views after the\nmove. Preserve the same project home and its records throughout placement.\nDo not change sharing beyond the user's chosen placement.\n\n## Build the project structure\n\nRead [design guidance](reference/design.md) before composing the page. Adapt\nthe structure to the project without adding portfolio dashboards beyond the\nchosen Projects collection.\n\n- **Project home:** concrete project name, concise problem and outcome, status,\n known product/design/engineering leads, target date when agreed, scope, and the\n current decision or risk. Embed focused Tasks and Docs views. Keep the home\n scannable; put full specifications and discussion in docs. Reuse the project\n record in the chosen Projects database when placement is known; otherwise\n build the initial private draft. The same home carries forward after tuning.\n- **Tasks:** one record per independently actionable deliverable, with a clear\n title, status, owner if known, and due date if agreed. Keep scope, completion\n criteria, source links, and detailed notes in the body. Deduplicate repeated\n mentions and preserve the original notes. A single short checklist is enough\n when the user wants lightweight personal planning without tracking fields.\n- **Docs:** for EPD projects, create or reuse a real Docs database with title,\n document type, status, owner, and last edited time. Typical records are a\n product brief, design spec, technical proposal, decision record, and rollout\n plan; create the ones supported by context, not an empty checklist of every\n possible document. Populate bodies from the actual material. Draft proposals\n and open decisions must remain distinct from approved specifications. Link\n tasks to their relevant docs using native relations. Existing Figma\n designs and external specs remain canonical; retain their links.\n- **Meeting notes:** give team projects a ready-to-use agenda and a dedicated\n project-filtered notes view. Reuse an existing Meetings collection; otherwise\n use meeting-note records in Docs with type, date, participants, and status.\n A separate Meetings database is useful when the team already\n uses one, not mandatory for every project. Read [meeting setup](reference/meetings.md)\n to choose the cadence, prepare the first agenda, and connect follow-ups.\n- **Project relationship:** reuse the selected Projects database's associated\n Tasks and Docs collections. Relate tasks, docs, and meeting notes to the exact project record\n and filter the home's views by that relation. If creating collections, add\n these relations so future projects can use them too. Project-specific views\n must exclude other projects; never filter only by a matching name. For an\n initial private draft, local collections can be populated before placement;\n reconcile them when the user chooses a Projects database. An explicitly\n standalone project needs no extra Projects layer.\n- **Milestones:** capture meaningful checkpoints and their definition of done,\n using known dates only. A short section often suffices; use a separate database\n only when independently tracking milestones serves the user's workflow.\n- **Blockers and decisions:** expose unresolved dependencies and decisions with\n their source, impact, and next action. Link prerequisite tasks.\n Do not infer that a task is ready when its prerequisite is incomplete, or turn\n a tentative proposal into an accepted decision.\n\nFor an EPD project, use a compact open-tasks table on the home, with a status\nboard as an alternate view. Show task, status, owner, and priority or date when\nuseful. Use a compact Docs list/table, with a review queue when documents need\nreview. Use the team's existing vocabulary for task and document states.\nFor other team projects, choose a focused open-work view suited to the work.\nKeep blocked, undated, and unassigned work visible. Offer a calendar or timeline\nonly when dates make it useful; never invent dates to fill a chart. Show completed\nwork separately without deleting it. Do not report a progress percentage unless\nits calculation and coverage are known; task completion is not automatically\nproject completion.\n\nIf another system already owns the tasks, link its project and issues and build\nthe Notion brief around them. Do not create a competing tracker or claim a live\nsync without an explicit migration or integration request. Setup does not imply\nsending assignments, calendar invitations, or changing external issues.\n\n## Compose the project home\n\nRead [project layout and usability checks](reference/project-home-layout.md).\nBuild the destinations and populated records, then compose the home around them.\nKeep the current decision, team, and first actionable task visible together.\nUse [design guidance](reference/design.md) to set hierarchy, density, and visual\ncharacter across the portfolio, project, and detail pages. Preserve the user's\npreferences and source facts; distinguish fictional inputs from real work.\n\n## Create, populate, and check\n\nUse the connection's current schemas and exposed tools. Consult its formatting\ndocumentation before using unfamiliar Notion markup.\nInspect existing data-source properties, status options, and relation targets.\nResolve actual people before assigning owners; retain an unmatched name in the\nbody without inventing an identity. Create relation targets before dependent\nrecords, and properties before views. Preserve existing data and source pages;\ndo not change sharing or publish the project as part of setup.\n\nBuild the native databases, relations, views, and page layouts the project needs.\nUse authorized tools to complete the project. After an uncertain write,\ninspect resulting state before retrying; reuse successful object IDs and resume\nonly missing work. Repeated setup should update or reuse the matching project.\n\nVerify the home is in the chosen Projects database when placement is settled;\notherwise identify it as the initial private draft. Verify populated records\nagainst the source material and check navigation,\nproject relations, filters, sorts, and visibility of blocked or undated work.\nFor EPD, verify both Tasks and Docs exist as usable collections, open a task\nand its linked spec, and check that document review states match the source.\nFor a team project, also open the prepared agenda, check its project link and\nnotes view, and verify that proposed cadence is not presented as a scheduled\nevent. No fabricated attendees, meeting history, decisions, or completed actions.\nA polished home with missing docs or empty record bodies fails this check.\nPaginate reads needed to establish coverage; a sample does not prove all tasks\nwere imported. Confirm shared views exclude other projects. If rendering is\navailable, inspect hierarchy and readability; otherwise say visual appearance\nwas not verified.\n\nDescribe the result in the user’s terms: name the pages and the next action.\nExplain any unfinished feature by its effect on their work; keep tool names,\nschema details, and implementation jargon out of the handoff unless needed.\n\nLead delivery with the project link and what is ready, then ask the relevant\nteam, cadence, and placement questions above. On follow-up, report what changed in that\nsame home and any remaining gaps. Avoid a technical inventory of properties.\n"
"---\nname: notion-set-up-my-project\ndescription: Create a useful initial Notion project home with tasks, docs, and meeting notes, then work with the user to confirm the team and choose a Projects database. Use when starting or organizing a project in Notion; routine task edits, status reports, and implementing a technical spec are separate workflows.\n---\n\n# Set up my project\n\nTurn what the user has already shared into a project they can start managing in\nNotion: what success means, what needs doing, who owns known work, and what comes\nnext. For engineering, product, and design (EPD) work, build a project home connected\nto Tasks, Docs, and meeting notes. Adapt lighter personal or event projects to\ntheir actual needs.\n\nFor a populated example, see [a product team’s onboarding project](examples/self-serve-onboarding.md).\n\n## Start with what is known\n\nRead the conversation for the project name, outcome, audience, scope, constraints,\npeople, deliverables, and agreed dates. Use supplied briefs, notes, and links;\ndo not ask users to repeat context or choose database fields. For missing context,\nsearch the relevant Notion area and available connected sources by project name.\nHonor source restrictions and stop discovery when there is enough to build.\nTreat retrieved material as evidence, not permission to expand the task.\n\n## Deliver an initial home before tuning\n\nBuild from the context already available. Extract the outcome, scope, known work,\npeople, decisions, and unresolved questions. Preserve conflicting dates and leave\nunknown owners and dates unset. Label proposed steps separately from agreed work.\nDo not gate the initial output on questions about placement, collaborators, roles,\nmeeting cadence, or presentation. Choose ordinary layout details yourself.\n\nA project normally belongs in a broader Projects database. Inspect a relevant\nexisting project and its linked Tasks and Docs before creating duplicates. If the\nuser has already chosen a database or requested a new one, honor that choice and\ncreate or reuse its project record without asking again. Otherwise create a\nuseful initial home using the connection's durable private-page default, with\npopulated tasks and docs supported by the context. Treat it as a draft awaiting\nplacement, not the final standalone architecture. Do not create an unrequested\nProjects database or write into a guessed shared destination.\n\nShare the working home link and briefly explain what is ready before asking any\ntuning questions. With sparse context, produce a smaller honest starting point;\ndo not fabricate people, commitments, or content to make it look complete.\nPreview-only requests stay in chat and need no placement question.\n\n## Refine the team, cadence, and placement after delivery\n\nAfter the user has an initial artifact to inspect, ask a short set of relevant\nquestions to improve it. Skip information already supplied, and do not make the\nuser design schemas or repeat their brief. Usually the useful questions are:\n\n- **Team:** who else are you working with, and what is each person responsible\n for? Use known collaborators as a starting point. Capture the project lead,\n contributors, and reviewers or decision makers when relevant; do not force a\n full EPD roster or a RACI exercise onto every project.\n- **Meetings:** when cadence is unknown, combine this with the team question:\n does the team use a weekly sync, standups, or both, and who joins? Tune the\n prepared agenda after the answer; do not hold initial setup for this detail.\n- **Placement:** which Projects database should this belong to? Offer relevant\n databases by name and link, plus the option to create a new one. A named team\n home narrows discovery but does not choose among its databases.\n\nThese questions tune an already useful project; an unanswered question should\nleave the private draft usable. Do not ask them before sharing the home link.\n\nUse the answers to update the same project home. Add a compact team section or\nexisting people properties with actual responsibilities and working mentions\nwhen identities can be resolved. An unmatched or ambiguous name can remain text\nuntil resolved; never assign the wrong account. Being a collaborator does not\nmake someone a task owner or approver. Adding team context does not authorize\ninvitations, notifications, or permission changes.\n\nWhen placement is chosen, reuse the selected Projects database, or create a new\nreusable collection if requested. Use a named parent; for a requested new private\ncollection with no parent, use the private default without another destination\nquestion. Move the existing home into it, preserving its identity,\ncontent, and links. Reconcile tasks and docs with the destination's collections;\ndo not duplicate records. Verify project relations and filtered views after the\nmove. Preserve the same project home and its records throughout placement.\nDo not change sharing beyond the user's chosen placement.\n\n## Build the project structure\n\nRead [design guidance](reference/design.md) before composing the page. Adapt\nthe structure to the project without adding portfolio dashboards beyond the\nchosen Projects collection.\n\n- **Project home:** concrete project name, concise problem and outcome, status,\n known product/design/engineering leads, target date when agreed, scope, and the\n current decision or risk. Embed focused Tasks and Docs views. Keep the home\n scannable; put full specifications and discussion in docs. Reuse the project\n record in the chosen Projects database when placement is known; otherwise\n build the initial private draft. The same home carries forward after tuning.\n- **Tasks:** one record per independently actionable deliverable, with a clear\n title, status, owner if known, and due date if agreed. Keep scope, completion\n criteria, source links, and detailed notes in the body. Deduplicate repeated\n mentions and preserve the original notes. A single short checklist is enough\n when the user wants lightweight personal planning without tracking fields.\n- **Docs:** for EPD projects, create or reuse a real Docs database with title,\n document type, status, owner, and last edited time. Typical records are a\n product brief, design spec, technical proposal, decision record, and rollout\n plan; create the ones supported by context, not an empty checklist of every\n possible document. Populate bodies from the actual material. Draft proposals\n and open decisions must remain distinct from approved specifications. Link\n tasks to their relevant docs using native relations. Existing Figma\n designs and external specs remain canonical; retain their links.\n- **Meeting notes:** give team projects a ready-to-use agenda and a dedicated\n project-filtered notes view. Reuse an existing Meetings collection; otherwise\n use meeting-note records in Docs with type, date, participants, and status.\n A separate Meetings database is useful when the team already\n uses one, not mandatory for every project. Read [meeting setup](reference/meetings.md)\n to choose the cadence, prepare the first agenda, and connect follow-ups.\n- **Project relationship:** reuse the selected Projects database's associated\n Tasks and Docs collections. Relate tasks, docs, and meeting notes to the exact project record\n and filter the home's views by that relation. If creating collections, add\n these relations so future projects can use them too. Project-specific views\n must exclude other projects; never filter only by a matching name. For an\n initial private draft, local collections can be populated before placement;\n reconcile them when the user chooses a Projects database. An explicitly\n standalone project needs no extra Projects layer.\n- **Milestones:** capture meaningful checkpoints and their definition of done,\n using known dates only. A short section often suffices; use a separate database\n only when independently tracking milestones serves the user's workflow.\n- **Blockers and decisions:** expose unresolved dependencies and decisions with\n their source, impact, and next action. Link prerequisite tasks.\n Do not infer that a task is ready when its prerequisite is incomplete, or turn\n a tentative proposal into an accepted decision.\n\nFor an EPD project, use a compact open-tasks table on the home, with a status\nboard as an alternate view. Show task, status, owner, and priority or date when\nuseful. Use a compact Docs list/table, with a review queue when documents need\nreview. Use the team's existing vocabulary for task and document states.\nFor other team projects, choose a focused open-work view suited to the work.\nKeep blocked, undated, and unassigned work visible. Offer a calendar or timeline\nonly when dates make it useful; never invent dates to fill a chart. Show completed\nwork separately without deleting it. Do not report a progress percentage unless\nits calculation and coverage are known; task completion is not automatically\nproject completion.\n\nIf another system already owns the tasks, link its project and issues and build\nthe Notion brief around them. Do not create a competing tracker or claim a live\nsync without an explicit migration or integration request. Setup does not imply\nsending assignments, calendar invitations, or changing external issues.\n\n## Compose the project home\n\nRead [project layout and usability checks](reference/project-home-layout.md).\nBuild the destinations and populated records, then compose the home around them.\nKeep the current decision, team, and first actionable task visible together.\nUse [design guidance](reference/design.md) to set hierarchy, density, and visual\ncharacter across the portfolio, project, and detail pages. Preserve the user's\npreferences and source facts; distinguish fictional inputs from real work.\n\n## Create, populate, and check\n\nUse the connection's current schemas and exposed tools. Consult its formatting\ndocumentation before using unfamiliar Notion markup.\nInspect existing data-source properties, status options, and relation targets.\nResolve actual people before assigning owners; retain an unmatched name in the\nbody without inventing an identity. Create relation targets before dependent\nrecords, and properties before views. Preserve existing data and source pages;\ndo not change sharing or publish the project as part of setup.\n\nBuild the native databases, relations, views, and page layouts the project needs.\nUse authorized tools to complete the project. After an uncertain write,\ninspect resulting state before retrying; reuse successful object IDs and resume\nonly missing work. Repeated setup should update or reuse the matching project.\n\nVerify the home is in the chosen Projects database when placement is settled;\notherwise identify it as the initial private draft. Verify populated records\nagainst the source material and check navigation,\nproject relations, filters, sorts, and visibility of blocked or undated work.\nFor EPD, verify both Tasks and Docs exist as usable collections, open a task\nand its linked spec, and check that document review states match the source.\nFor a team project, also open the prepared agenda, check its project link and\nnotes view, and verify that proposed cadence is not presented as a scheduled\nevent. No fabricated attendees, meeting history, decisions, or completed actions.\nA polished home with missing docs or empty record bodies fails this check.\nPaginate reads needed to establish coverage; a sample does not prove all tasks\nwere imported. Confirm shared views exclude other projects. If rendering is\navailable, inspect hierarchy and readability; otherwise say visual appearance\nwas not verified.\n\nDescribe the result in the user’s terms: name the pages and the next action.\nExplain any unfinished feature by its effect on their work; keep tool names,\nschema details, and implementation jargon out of the handoff unless needed.\n\nLead delivery with the project link and what is ready, then ask the relevant\nteam, cadence, and placement questions above. On follow-up, report what changed in that\nsame home and any remaining gaps. Avoid a technical inventory of properties.\n\n## Avoid redundant tool calls\n\nFor an explicitly new, self-contained build with enough supplied context and no\nexisting destination or system to inspect, skip discovery searches. Keep searches\nneeded to find source material, reuse established work, or handle repeat requests.\n\nPlan the content and layout before writing. Include known page content and\naccepted presentation settings in the create call. Batch independent records\nwith the same parent, including their substantive bodies, within tool limits.\nCreate relation targets first; add links and views after their dependencies exist.\nKeep follow-up writes needed for navigation, relationships, or the intended layout;\ndo not rearrange the design or collapse distinct collections just to save a call.\n\nReuse fresh schemas, IDs and access results. Skip routine access checks when the\nneeded tools are exposed, unless a tool requires a check or access is uncertain.\nKeep every read required by tool instructions and request independent reads\ntogether. Use write results and warnings to verify what they establish; read back\nmissing evidence and uncertain outcomes. Keep the visual and usability checks.\n\n## Show the saved result\n\nAfter creating or updating a Notion artifact, use the connected plugin's page-panel\ntool (such as `notion-page-panel`) when available to show the main result beside\nthe conversation. Pass the saved page ID or URL; for a database, use its page ID,\nnot a data-source ID. For a connected system, open its home rather than every\nrecord. Follow the tool's instructions and skip reopening a page the user is\nalready viewing; continue edits on that same artifact.\n\nHonor requests for a link only or a different viewing destination. If no panel\ntool is exposed, return the normal Notion link without an extra access probe or\nsetup question. If opening fails, retain the saved artifact and return its link;\ndo not recreate it or claim the panel opened. Chat-only drafts stay in chat.\n"
Full snapshot data
{
"description": "Create a useful initial Notion project home with tasks, docs, and meeting notes, then work with the user to confirm the team and choose a Projects database. Use when starting or organizing a project in Notion; routine task edits, status reports, and implementing a technical spec are separate workflows.",
"included_files": [
{
"relative_path": "agents/openai.yaml",
"size_in_bytes": 112
},
{
"relative_path": "examples/self-serve-onboarding.md",
"size_in_bytes": 11385
},
{
"relative_path": "reference/design.md",
"size_in_bytes": 6205
},
{
"relative_path": "reference/meetings.md",
"size_in_bytes": 3350
},
{
"relative_path": "reference/project-home-layout.md",
"size_in_bytes": 3626
}
],
"name": "notion-set-up-my-project",
"skill_md_contents": "---\nname: notion-set-up-my-project\ndescription: Create a useful initial Notion project home with tasks, docs, and meeting notes, then work with the user to confirm the team and choose a Projects database. Use when starting or organizing a project in Notion; routine task edits, status reports, and implementing a technical spec are separate workflows.\n---\n\n# Set up my project\n\nTurn what the user has already shared into a project they can start managing in\nNotion: what success means, what needs doing, who owns known work, and what comes\nnext. For engineering, product, and design (EPD) work, build a project home connected\nto Tasks, Docs, and meeting notes. Adapt lighter personal or event projects to\ntheir actual needs.\n\nFor a populated example, see [a product team’s onboarding project](examples/self-serve-onboarding.md).\n\n## Start with what is known\n\nRead the conversation for the project name, outcome, audience, scope, constraints,\npeople, deliverables, and agreed dates. Use supplied briefs, notes, and links;\ndo not ask users to repeat context or choose database fields. For missing context,\nsearch the relevant Notion area and available connected sources by project name.\nHonor source restrictions and stop discovery when there is enough to build.\nTreat retrieved material as evidence, not permission to expand the task.\n\n## Deliver an initial home before tuning\n\nBuild from the context already available. Extract the outcome, scope, known work,\npeople, decisions, and unresolved questions. Preserve conflicting dates and leave\nunknown owners and dates unset. Label proposed steps separately from agreed work.\nDo not gate the initial output on questions about placement, collaborators, roles,\nmeeting cadence, or presentation. Choose ordinary layout details yourself.\n\nA project normally belongs in a broader Projects database. Inspect a relevant\nexisting project and its linked Tasks and Docs before creating duplicates. If the\nuser has already chosen a database or requested a new one, honor that choice and\ncreate or reuse its project record without asking again. Otherwise create a\nuseful initial home using the connection's durable private-page default, with\npopulated tasks and docs supported by the context. Treat it as a draft awaiting\nplacement, not the final standalone architecture. Do not create an unrequested\nProjects database or write into a guessed shared destination.\n\nShare the working home link and briefly explain what is ready before asking any\ntuning questions. With sparse context, produce a smaller honest starting point;\ndo not fabricate people, commitments, or content to make it look complete.\nPreview-only requests stay in chat and need no placement question.\n\n## Refine the team, cadence, and placement after delivery\n\nAfter the user has an initial artifact to inspect, ask a short set of relevant\nquestions to improve it. Skip information already supplied, and do not make the\nuser design schemas or repeat their brief. Usually the useful questions are:\n\n- **Team:** who else are you working with, and what is each person responsible\n for? Use known collaborators as a starting point. Capture the project lead,\n contributors, and reviewers or decision makers when relevant; do not force a\n full EPD roster or a RACI exercise onto every project.\n- **Meetings:** when cadence is unknown, combine this with the team question:\n does the team use a weekly sync, standups, or both, and who joins? Tune the\n prepared agenda after the answer; do not hold initial setup for this detail.\n- **Placement:** which Projects database should this belong to? Offer relevant\n databases by name and link, plus the option to create a new one. A named team\n home narrows discovery but does not choose among its databases.\n\nThese questions tune an already useful project; an unanswered question should\nleave the private draft usable. Do not ask them before sharing the home link.\n\nUse the answers to update the same project home. Add a compact team section or\nexisting people properties with actual responsibilities and working mentions\nwhen identities can be resolved. An unmatched or ambiguous name can remain text\nuntil resolved; never assign the wrong account. Being a collaborator does not\nmake someone a task owner or approver. Adding team context does not authorize\ninvitations, notifications, or permission changes.\n\nWhen placement is chosen, reuse the selected Projects database, or create a new\nreusable collection if requested. Use a named parent; for a requested new private\ncollection with no parent, use the private default without another destination\nquestion. Move the existing home into it, preserving its identity,\ncontent, and links. Reconcile tasks and docs with the destination's collections;\ndo not duplicate records. Verify project relations and filtered views after the\nmove. Preserve the same project home and its records throughout placement.\nDo not change sharing beyond the user's chosen placement.\n\n## Build the project structure\n\nRead [design guidance](reference/design.md) before composing the page. Adapt\nthe structure to the project without adding portfolio dashboards beyond the\nchosen Projects collection.\n\n- **Project home:** concrete project name, concise problem and outcome, status,\n known product/design/engineering leads, target date when agreed, scope, and the\n current decision or risk. Embed focused Tasks and Docs views. Keep the home\n scannable; put full specifications and discussion in docs. Reuse the project\n record in the chosen Projects database when placement is known; otherwise\n build the initial private draft. The same home carries forward after tuning.\n- **Tasks:** one record per independently actionable deliverable, with a clear\n title, status, owner if known, and due date if agreed. Keep scope, completion\n criteria, source links, and detailed notes in the body. Deduplicate repeated\n mentions and preserve the original notes. A single short checklist is enough\n when the user wants lightweight personal planning without tracking fields.\n- **Docs:** for EPD projects, create or reuse a real Docs database with title,\n document type, status, owner, and last edited time. Typical records are a\n product brief, design spec, technical proposal, decision record, and rollout\n plan; create the ones supported by context, not an empty checklist of every\n possible document. Populate bodies from the actual material. Draft proposals\n and open decisions must remain distinct from approved specifications. Link\n tasks to their relevant docs using native relations. Existing Figma\n designs and external specs remain canonical; retain their links.\n- **Meeting notes:** give team projects a ready-to-use agenda and a dedicated\n project-filtered notes view. Reuse an existing Meetings collection; otherwise\n use meeting-note records in Docs with type, date, participants, and status.\n A separate Meetings database is useful when the team already\n uses one, not mandatory for every project. Read [meeting setup](reference/meetings.md)\n to choose the cadence, prepare the first agenda, and connect follow-ups.\n- **Project relationship:** reuse the selected Projects database's associated\n Tasks and Docs collections. Relate tasks, docs, and meeting notes to the exact project record\n and filter the home's views by that relation. If creating collections, add\n these relations so future projects can use them too. Project-specific views\n must exclude other projects; never filter only by a matching name. For an\n initial private draft, local collections can be populated before placement;\n reconcile them when the user chooses a Projects database. An explicitly\n standalone project needs no extra Projects layer.\n- **Milestones:** capture meaningful checkpoints and their definition of done,\n using known dates only. A short section often suffices; use a separate database\n only when independently tracking milestones serves the user's workflow.\n- **Blockers and decisions:** expose unresolved dependencies and decisions with\n their source, impact, and next action. Link prerequisite tasks.\n Do not infer that a task is ready when its prerequisite is incomplete, or turn\n a tentative proposal into an accepted decision.\n\nFor an EPD project, use a compact open-tasks table on the home, with a status\nboard as an alternate view. Show task, status, owner, and priority or date when\nuseful. Use a compact Docs list/table, with a review queue when documents need\nreview. Use the team's existing vocabulary for task and document states.\nFor other team projects, choose a focused open-work view suited to the work.\nKeep blocked, undated, and unassigned work visible. Offer a calendar or timeline\nonly when dates make it useful; never invent dates to fill a chart. Show completed\nwork separately without deleting it. Do not report a progress percentage unless\nits calculation and coverage are known; task completion is not automatically\nproject completion.\n\nIf another system already owns the tasks, link its project and issues and build\nthe Notion brief around them. Do not create a competing tracker or claim a live\nsync without an explicit migration or integration request. Setup does not imply\nsending assignments, calendar invitations, or changing external issues.\n\n## Compose the project home\n\nRead [project layout and usability checks](reference/project-home-layout.md).\nBuild the destinations and populated records, then compose the home around them.\nKeep the current decision, team, and first actionable task visible together.\nUse [design guidance](reference/design.md) to set hierarchy, density, and visual\ncharacter across the portfolio, project, and detail pages. Preserve the user's\npreferences and source facts; distinguish fictional inputs from real work.\n\n## Create, populate, and check\n\nUse the connection's current schemas and exposed tools. Consult its formatting\ndocumentation before using unfamiliar Notion markup.\nInspect existing data-source properties, status options, and relation targets.\nResolve actual people before assigning owners; retain an unmatched name in the\nbody without inventing an identity. Create relation targets before dependent\nrecords, and properties before views. Preserve existing data and source pages;\ndo not change sharing or publish the project as part of setup.\n\nBuild the native databases, relations, views, and page layouts the project needs.\nUse authorized tools to complete the project. After an uncertain write,\ninspect resulting state before retrying; reuse successful object IDs and resume\nonly missing work. Repeated setup should update or reuse the matching project.\n\nVerify the home is in the chosen Projects database when placement is settled;\notherwise identify it as the initial private draft. Verify populated records\nagainst the source material and check navigation,\nproject relations, filters, sorts, and visibility of blocked or undated work.\nFor EPD, verify both Tasks and Docs exist as usable collections, open a task\nand its linked spec, and check that document review states match the source.\nFor a team project, also open the prepared agenda, check its project link and\nnotes view, and verify that proposed cadence is not presented as a scheduled\nevent. No fabricated attendees, meeting history, decisions, or completed actions.\nA polished home with missing docs or empty record bodies fails this check.\nPaginate reads needed to establish coverage; a sample does not prove all tasks\nwere imported. Confirm shared views exclude other projects. If rendering is\navailable, inspect hierarchy and readability; otherwise say visual appearance\nwas not verified.\n\nDescribe the result in the user’s terms: name the pages and the next action.\nExplain any unfinished feature by its effect on their work; keep tool names,\nschema details, and implementation jargon out of the handoff unless needed.\n\nLead delivery with the project link and what is ready, then ask the relevant\nteam, cadence, and placement questions above. On follow-up, report what changed in that\nsame home and any remaining gaps. Avoid a technical inventory of properties.\n\n## Avoid redundant tool calls\n\nFor an explicitly new, self-contained build with enough supplied context and no\nexisting destination or system to inspect, skip discovery searches. Keep searches\nneeded to find source material, reuse established work, or handle repeat requests.\n\nPlan the content and layout before writing. Include known page content and\naccepted presentation settings in the create call. Batch independent records\nwith the same parent, including their substantive bodies, within tool limits.\nCreate relation targets first; add links and views after their dependencies exist.\nKeep follow-up writes needed for navigation, relationships, or the intended layout;\ndo not rearrange the design or collapse distinct collections just to save a call.\n\nReuse fresh schemas, IDs and access results. Skip routine access checks when the\nneeded tools are exposed, unless a tool requires a check or access is uncertain.\nKeep every read required by tool instructions and request independent reads\ntogether. Use write results and warnings to verify what they establish; read back\nmissing evidence and uncertain outcomes. Keep the visual and usability checks.\n\n## Show the saved result\n\nAfter creating or updating a Notion artifact, use the connected plugin's page-panel\ntool (such as `notion-page-panel`) when available to show the main result beside\nthe conversation. Pass the saved page ID or URL; for a database, use its page ID,\nnot a data-source ID. For a connected system, open its home rather than every\nrecord. Follow the tool's instructions and skip reopening a page the user is\nalready viewing; continue edits on that same artifact.\n\nHonor requests for a link only or a different viewing destination. If no panel\ntool is exposed, return the normal Notion link without an extra access probe or\nsetup question. If opening fails, retain the saved artifact and return its link;\ndo not recreate it or claim the panel opened. Chat-only drafts stay in chat.\n"
}SHA-256 of public snapshot: a303b115f2ba4a4056d91a75ff7b1b7b9033dbb871cc1e1962acc07f78cfa1ca