[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6MGWP87632DF426S8PW5ZNN":3,"skill-same:01M6MGWP87632DF426S8PW5ZNN":80},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":45,"linkOnly":46,"markdownContent":47,"markdownBody":48,"frontMatter":49,"refName":50,"commitSha":51,"skillDirectory":52,"forksCount":53,"repositoryHtmlUrl":54,"repositoryHomepageUrl":55,"ownerHtmlUrl":56,"ownerBlogUrl":57,"ownerDescription":58,"updatedAt":59,"install":60,"resources":73},"01M6MGWP87632DF426S8PW5ZNN","context-item-creation",null,"Guides you through creating context items — instructions, skills, knowledge pointers, and memory events — that give the AI analyst persistent knowledge about your organization. Covers when to create each type, naming conventions, folder organization, and the hard boundary between context (rules and procedures) and the semantic layer (logic and calculations).","hashgraph-online\u002Fawesome-codex-plugins","hashgraph-online","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F160371962?v=4","01M6CNB6NRJVNNGN14TR01Q6D3","HOL (Hashgraph Online)","01M6JT1HSKKG9V68KQAQGBA7XP",1575,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fhoneydew-ai\u002Fhoneydew-ai-coding-agents-plugins\u002Fskills\u002Fcontext-item-creation\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44],"context item creation","item creation","context","creation","when","memory","layer","folder","you","naming","guides","events","type","logic","analyst","organization","instructions","Rules","knowledge","boundary","create","2026-11-10 10:12:30",false,"---\nname: context-item-creation\ndescription: Guides you through creating context items — instructions, skills, knowledge pointers, and memory events — that give the AI analyst persistent knowledge about your organization. Covers when to create each type, naming conventions, folder organization, and the hard boundary between context (rules and procedures) and the semantic layer (logic and calculations).\n---\n\n## Prerequisites\n\nBefore creating context items, ensure you are on the correct workspace and branch. Use `get_session_workspace_and_branch` to check the current session context. For development work, create a branch with `create_workspace_branch` (the session switches automatically). See the `model-exploration` skill for the full workspace\u002Fbranch tool reference.\n\n---\n\n## Overview\n\n**Context items** are the mechanism for giving the AI analyst persistent, organizational knowledge: standing rules it must always follow, on-demand procedures it can retrieve, external references it can look up, and a historical record of notable events.\n\nContext items belong to the **context layer** — they are consumed by the Honeydew AI analyst at query time to shape its behavior. They are separate from the **semantic layer** (entities, metrics, attributes, relations, domains), which defines your data model and business logic such as metric calculations.\n\nThere are four types:\n\n| Type | Subtype | When retrieved | Use for |\n|---|---|---|---|\n| **Instruction** | `instruction` | Every prompt, automatically | Short standing rules or constraints the model must always respect |\n| **Skill** | `skill` | On demand, when relevant | Long-form analysis playbooks and procedural guides |\n| **Knowledge** | `knowledge` | On demand, when relevant | Pointers to external content (Confluence, Notion, etc.) |\n| **Memory** | `event` | On demand, when relevant | Historical events and decisions that affect data interpretation |\n\n> **Before creating anything, read the critical boundary section below.** Most requests for context items are actually requests to define or document logic — which belongs in the semantic layer, not in context.\n\n---\n\n## THE HARD BOUNDARY: Context vs. Semantic Layer\n\n> **This is the most important rule in this skill. Re-read it every time you are asked to create a context item.**\n\n**Context items are for rules, procedures, and history. They are NOT for logic, calculations, or data definitions.**\n\nIf the information describes *how to compute something*, *what a value means mathematically*, or *how data is structured* — it belongs in the **semantic layer** (as a metric, attribute, filter, relationship, or entity definition), not in a context item.\n\n### When to use the Semantic Layer instead\n\n| Request | ❌ Wrong approach | ✅ Correct approach |\n|---|---|---|\n| \"Add a rule that revenue = price × quantity\" | Instruction: \"revenue is price × quantity\" | Calculated attribute or metric with `sql: orders.price * orders.quantity` |\n| \"Explain how to calculate LTV\" | Instruction with formula | Metric with the LTV SQL expression |\n| \"Document that active customers are those with an order in 90 days\" | Instruction with definition | Calculated attribute `is_active` with `sql: orders.last_order_date >= CURRENT_DATE - 90` |\n| \"Add a rule that we always filter to the current fiscal year\" | Instruction with filter logic | Domain filter or global parameter — *unless* the preference is about which existing filter to apply, in which case an instruction referencing the filter by name is fine |\n| \"Record that the orders table has columns order_id, customer_id...\" | Knowledge item | Entity and dataset definition via `entity-creation` skill |\n| \"Add a note that revenue excludes returns\" | Instruction | Metric `sql` with `FILTER (WHERE NOT orders.is_return)`, plus a clear `description` on the metric |\n\n### What context items ARE good for\n\n- **Instructions**: A standing preference that refers to existing semantic objects, not logic. \"Always use `orders.net_revenue` rather than `orders.gross_revenue` for revenue questions.\"\n- **Skills**: A multi-step analytical procedure: \"To investigate churn, start by segmenting cohorts, then compare retention curves, then identify the inflection point.\"\n- **Knowledge**: A pointer to external policy documents, runbooks, or reference material.\n- **Memory\u002FEvents**: \"On 2024-03-01 we redefined revenue to exclude refunds — historical data was backfilled.\"\n\n**The test:** Can you express this as a SQL expression or a schema definition? If yes, it belongs in the semantic layer. Context items should reference semantic objects by name, not define or duplicate their logic.\n\n---\n\n## MCP Tools\n\n### create_context_item\n\nCall `create_context_item` with the frontmatter fields as `context_item` parameters and the prose body as `markdown_text`.\n\n### After Creation\u002FUpdate: Display the UI Link\n\nAfter a successful `create_context_item` or `update_context_item` call, the response includes a `ui_url` field. **Always display this URL to the user** so they can quickly open the item in the Honeydew application.\n\n### update_context_item\n\nTo modify an existing context item:\n\n1. Call `get_context_item` with the item's name to read its current content.\n2. Call `update_context_item` with the item's `name` and the updated `context_item` fields and `markdown_text`.\n\n> **Minimal diff rule:** Only change the fields you need to modify. Context items are versioned in git, so unnecessary reformatting creates noisy diffs.\n\n### delete_context_item\n\nCall `delete_context_item` with the item's `name`.\n\n| Task | Tool |\n|---|---|\n| Create | `create_context_item` |\n| Update | `update_context_item` |\n| Delete | `delete_context_item` |\n| Get one | `get_context_item` |\n| List all | `list_context_items` |\n\n---\n\n## Naming and Organization\n\n### Name format\n\nNames use `folder\u002Fitem-name` with lowercase kebab-case and `\u002F` as a hierarchy separator:\n\n```\nfinance\u002Fuse-net-revenue\ncustomer\u002Fchurn-analysis-playbook\nconfluence\u002Fdata-governance-policy\nfinance\u002Frevenue-redefinition-2024\n```\n\nA flat name (no folder) is fine when the item is genuinely cross-cutting:\n\n```\nroot-cause-analysis\ngdpr-compliance-rollout\n```\n\n### Why folders matter — agent scoping\n\nWhen configuring an AI agent, context items are referenced using **glob patterns**. Folders let you include a coherent group of items with a single pattern:\n\n- `finance\u002F*` — all finance-specific rules and skills\n- `customer\u002F*` — all customer analysis guidelines\n- `*\u002Fchurn-*` — all churn-related items regardless of team folder\n- `menu\u002F**` — all items under `menu\u002F` including subfolders (e.g. `menu\u002Fdrinks\u002Flatte-art`)\n- `**` — all context items across all folders and subfolders (use sparingly; typically for a catch-all agent)\n\n**Organize by the team or domain that owns the context**, not by the item type. Items of different types (instruction + skill + knowledge) for the same domain belong in the same folder:\n\n```\nfinance\u002Fuse-net-revenue         ← instruction\nfinance\u002Fvariance-analysis       ← skill\nfinance\u002Faccounting-policy       ← knowledge\nfinance\u002Frevenue-redefinition    ← memory\n```\n\nThis way, an agent configured with `finance\u002F*` automatically gets all finance context.\n\n### Naming best practices\n\n- Use the folder to reflect ownership, not to repeat the item type. `finance\u002Fuse-net-revenue` is good. `instructions\u002Fuse-net-revenue` is redundant.\n- Name after the concept or rule, not the implementation. `exclude-canceled-orders` is better than `filter-status-not-canceled`.\n- For memory events, name after the event itself, not a generic label. `revenue-redefinition` or `gdpr-rollout` beats `data-change-3`.\n\n---\n\n## Frontmatter format\n\nContext items are stored as frontmatter documents: a YAML block between `---` delimiters, followed by a markdown prose body. When calling `create_context_item`, pass the YAML fields as `context_item` parameters and the prose body as the `markdown_text` parameter.\n\nThe `description` field is the **retrieval signal** — the model reads it to decide whether to fetch an on-demand item. It is **used for skills, knowledge, and memory events**. Do not add `description` to instructions; their only content is the prose body.\n\nSee [examples.md](examples.md) for what the full frontmatter documents look like.\n\n---\n\n## Type Reference\n\n### Instruction — standing rule (always on)\n\n**Use when:** There is a standing preference, constraint, or fact that the model must respect in every conversation. These are injected into every prompt automatically.\n\n**Keep instructions concise.** They are never filtered out, so verbose instructions bloat every prompt. If the content is more than a short paragraph, turn it into a skill instead.\n\n**Parameters:**\n- `type`: `instruction`\n- `subtype`: `instruction`\n- `name`: `folder\u002Frule-name`\n- `title`: short human label\n- prose body: **the rule text** — what the model must do or avoid\n- `related_objects`: *(optional)* only the fields or entities directly named or constrained by this instruction — not downstream consumers, not lineage, not entities that happen to join to a relevant one; use an entity reference instead of a field list when the instruction applies to most fields of that entity\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller; only set if a different team or person should be listed\n\n**Good instruction topics:**\n- Metric selection preferences (\"always use net_revenue\")\n- Domain-specific constraints (\"exclude test accounts from all analyses\")\n- Terminology corrections (\"what users call 'bookings' maps to the `orders.gmv` metric\")\n\n### Skill — on-demand playbook\n\n**Use when:** There is a multi-step analysis procedure the model should follow when asked about a specific topic. Skills are retrieved when the model judges them relevant based on the description.\n\n**Write a rich, keyword-dense `description`** — this is what the model reads to decide whether to retrieve the skill. A weak description means the skill will be missed.\n\n**Parameters:**\n- `type`: `instruction`\n- `subtype`: `skill`\n- `name`: `topic-or-folder\u002Ftopic`\n- `title`: short human label\n- `description`: **required** — precise, keyword-rich description of when to use this skill and what it covers (this is the retrieval signal)\n- prose body: **the playbook** — step-by-step guide written in markdown (goes after the closing `---`)\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller\n\n**Good skill topics:**\n- Root cause analysis procedures\n- Customer cohort or churn investigation workflows\n- Reporting preparation checklists\n- Domain-specific investigation playbooks\n\n### Knowledge — external source pointer\n\n**Use when:** There is an external document (Confluence page, Notion database, runbook, policy) that the model should consult when answering relevant questions. The model fetches the content via an MCP server at retrieval time.\n\n**Write a precise `description`** that includes the topics the document covers — this is the retrieval signal.\n\n**Parameters:**\n- `type`: `instruction`\n- `subtype`: `knowledge`\n- `name`: `source-folder\u002Fdocument-name`\n- `title`: short human label\n- `description`: **required** — what topics this document covers (used for retrieval)\n- `external_source.tool`: the MCP server name (e.g. `confluence`, `notion`)\n- `external_source.resource_id`: the URI within that server (e.g. `confluence:\u002F\u002FPage-Title`)\n- prose body: typically omit — the content is fetched from the external source at runtime\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller\n\n### Memory \u002F Event — historical record\n\n**Use when:** Something happened that affects how data should be interpreted: a metric was redefined, a process changed, a data migration occurred, a business event shifted baseline numbers.\n\nMemory items are retrieved on demand when the model judges them relevant to the current question.\n\n**Parameters:**\n- `type`: `memory`\n- `subtype`: `event`\n- `name`: `domain\u002Fevent-name`\n- `title`: short human label\n- `description`: **required** — what this event is about (used for retrieval)\n- prose body: **what happened and why it matters** for data interpretation\n- `from_date`: when the event occurred (point-in-time) or when it started (if a duration)\n- `to_date`: *(optional)* end date for events spanning a period\n- `related_objects`: *(optional)* only the fields or entities directly changed or affected by this event — not downstream consumers, not lineage, not entities that happen to join to a relevant one; use an entity reference instead of a field list when the event affects most fields of that entity\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller\n\n**Good memory topics:**\n- Metric or definition changes with a specific effective date\n- ETL migrations or data backfills\n- Significant business events (acquisitions, product launches) that cause discontinuities in data\n- Changes in how data is collected or tagged\n\n---\n\n## Examples\n\nSee [examples.md](examples.md) for full worked examples of each type, plus right-vs-wrong comparisons.\n\n---\n\n## Discovery Helpers\n\n- `list_context_items` — see all existing items (avoid duplicates)\n- `get_context_item` — read a specific item's current content before updating\n- `list_agents` \u002F `get_agent` — see which agents reference which context globs, to understand the scoping impact of adding a new item\n\n---\n\n## Documentation Lookup\n\nUse the `honeydew-docs` MCP tools to search the Honeydew documentation when:\n\n- You need to understand how context items interact with agents or how retrieval works\n- The user asks about agent configuration, glob patterns, or context scoping\n- You need guidance on memory subtypes or advanced context item options\n\nSearch for topics like: \"context items\", \"instructions\", \"agent context\", \"memory events\", \"skills\".\n\n---\n\n## Best Practices\n\n1. **Semantic layer first.** If the request is about logic, calculations, definitions, or data structure — build it in the semantic layer. Context items reference semantic objects; they do not replace them.\n\n2. **Instructions must be short.** They're injected into every prompt. If it doesn't fit in 2–3 sentences, turn it into a skill.\n\n3. **Skill descriptions are the retrieval key.** Write rich, specific, keyword-heavy descriptions. \"Guide for root cause analysis of revenue drops, covering cohort isolation, period-over-period comparison, and funnel decomposition\" is far better than \"revenue analysis\".\n\n4. **Folder by domain, not by type.** `finance\u002F` should contain instructions, skills, knowledge, and events all related to finance. This makes agent glob patterns meaningful.\n\n5. **Check for duplicates before creating.** Run `list_context_items` first. Overlapping or contradictory instructions are worse than none.\n\n6. **Use labels for cross-cutting concerns.** Labels like `pii`, `finance`, or `deprecated` enable filtering that cuts across folder hierarchy.\n\n---\n\n## MANDATORY: Confirm You Are Not Duplicating Semantic Layer Logic\n\nBefore creating any context item, verify:\n\n1. **Does this describe a calculation or data definition?** → Build it as a metric, attribute, or filter instead.\n2. **Does this duplicate something already in the semantic layer?** → Update the semantic object's `description` field instead of adding context.\n3. **Is this a preference that references an existing metric by name?** → Good candidate for an instruction.\n\nIf you are unsure, ask the user: *\"Should I add this as a rule for the AI analyst, or should I encode it directly in the semantic layer as a metric\u002Fattribute?\"*\n","## Prerequisites\n\nBefore creating context items, ensure you are on the correct workspace and branch. Use `get_session_workspace_and_branch` to check the current session context. For development work, create a branch with `create_workspace_branch` (the session switches automatically). See the `model-exploration` skill for the full workspace\u002Fbranch tool reference.\n\n---\n\n## Overview\n\n**Context items** are the mechanism for giving the AI analyst persistent, organizational knowledge: standing rules it must always follow, on-demand procedures it can retrieve, external references it can look up, and a historical record of notable events.\n\nContext items belong to the **context layer** — they are consumed by the Honeydew AI analyst at query time to shape its behavior. They are separate from the **semantic layer** (entities, metrics, attributes, relations, domains), which defines your data model and business logic such as metric calculations.\n\nThere are four types:\n\n| Type | Subtype | When retrieved | Use for |\n|---|---|---|---|\n| **Instruction** | `instruction` | Every prompt, automatically | Short standing rules or constraints the model must always respect |\n| **Skill** | `skill` | On demand, when relevant | Long-form analysis playbooks and procedural guides |\n| **Knowledge** | `knowledge` | On demand, when relevant | Pointers to external content (Confluence, Notion, etc.) |\n| **Memory** | `event` | On demand, when relevant | Historical events and decisions that affect data interpretation |\n\n> **Before creating anything, read the critical boundary section below.** Most requests for context items are actually requests to define or document logic — which belongs in the semantic layer, not in context.\n\n---\n\n## THE HARD BOUNDARY: Context vs. Semantic Layer\n\n> **This is the most important rule in this skill. Re-read it every time you are asked to create a context item.**\n\n**Context items are for rules, procedures, and history. They are NOT for logic, calculations, or data definitions.**\n\nIf the information describes *how to compute something*, *what a value means mathematically*, or *how data is structured* — it belongs in the **semantic layer** (as a metric, attribute, filter, relationship, or entity definition), not in a context item.\n\n### When to use the Semantic Layer instead\n\n| Request | ❌ Wrong approach | ✅ Correct approach |\n|---|---|---|\n| \"Add a rule that revenue = price × quantity\" | Instruction: \"revenue is price × quantity\" | Calculated attribute or metric with `sql: orders.price * orders.quantity` |\n| \"Explain how to calculate LTV\" | Instruction with formula | Metric with the LTV SQL expression |\n| \"Document that active customers are those with an order in 90 days\" | Instruction with definition | Calculated attribute `is_active` with `sql: orders.last_order_date >= CURRENT_DATE - 90` |\n| \"Add a rule that we always filter to the current fiscal year\" | Instruction with filter logic | Domain filter or global parameter — *unless* the preference is about which existing filter to apply, in which case an instruction referencing the filter by name is fine |\n| \"Record that the orders table has columns order_id, customer_id...\" | Knowledge item | Entity and dataset definition via `entity-creation` skill |\n| \"Add a note that revenue excludes returns\" | Instruction | Metric `sql` with `FILTER (WHERE NOT orders.is_return)`, plus a clear `description` on the metric |\n\n### What context items ARE good for\n\n- **Instructions**: A standing preference that refers to existing semantic objects, not logic. \"Always use `orders.net_revenue` rather than `orders.gross_revenue` for revenue questions.\"\n- **Skills**: A multi-step analytical procedure: \"To investigate churn, start by segmenting cohorts, then compare retention curves, then identify the inflection point.\"\n- **Knowledge**: A pointer to external policy documents, runbooks, or reference material.\n- **Memory\u002FEvents**: \"On 2024-03-01 we redefined revenue to exclude refunds — historical data was backfilled.\"\n\n**The test:** Can you express this as a SQL expression or a schema definition? If yes, it belongs in the semantic layer. Context items should reference semantic objects by name, not define or duplicate their logic.\n\n---\n\n## MCP Tools\n\n### create_context_item\n\nCall `create_context_item` with the frontmatter fields as `context_item` parameters and the prose body as `markdown_text`.\n\n### After Creation\u002FUpdate: Display the UI Link\n\nAfter a successful `create_context_item` or `update_context_item` call, the response includes a `ui_url` field. **Always display this URL to the user** so they can quickly open the item in the Honeydew application.\n\n### update_context_item\n\nTo modify an existing context item:\n\n1. Call `get_context_item` with the item's name to read its current content.\n2. Call `update_context_item` with the item's `name` and the updated `context_item` fields and `markdown_text`.\n\n> **Minimal diff rule:** Only change the fields you need to modify. Context items are versioned in git, so unnecessary reformatting creates noisy diffs.\n\n### delete_context_item\n\nCall `delete_context_item` with the item's `name`.\n\n| Task | Tool |\n|---|---|\n| Create | `create_context_item` |\n| Update | `update_context_item` |\n| Delete | `delete_context_item` |\n| Get one | `get_context_item` |\n| List all | `list_context_items` |\n\n---\n\n## Naming and Organization\n\n### Name format\n\nNames use `folder\u002Fitem-name` with lowercase kebab-case and `\u002F` as a hierarchy separator:\n\n```\nfinance\u002Fuse-net-revenue\ncustomer\u002Fchurn-analysis-playbook\nconfluence\u002Fdata-governance-policy\nfinance\u002Frevenue-redefinition-2024\n```\n\nA flat name (no folder) is fine when the item is genuinely cross-cutting:\n\n```\nroot-cause-analysis\ngdpr-compliance-rollout\n```\n\n### Why folders matter — agent scoping\n\nWhen configuring an AI agent, context items are referenced using **glob patterns**. Folders let you include a coherent group of items with a single pattern:\n\n- `finance\u002F*` — all finance-specific rules and skills\n- `customer\u002F*` — all customer analysis guidelines\n- `*\u002Fchurn-*` — all churn-related items regardless of team folder\n- `menu\u002F**` — all items under `menu\u002F` including subfolders (e.g. `menu\u002Fdrinks\u002Flatte-art`)\n- `**` — all context items across all folders and subfolders (use sparingly; typically for a catch-all agent)\n\n**Organize by the team or domain that owns the context**, not by the item type. Items of different types (instruction + skill + knowledge) for the same domain belong in the same folder:\n\n```\nfinance\u002Fuse-net-revenue         ← instruction\nfinance\u002Fvariance-analysis       ← skill\nfinance\u002Faccounting-policy       ← knowledge\nfinance\u002Frevenue-redefinition    ← memory\n```\n\nThis way, an agent configured with `finance\u002F*` automatically gets all finance context.\n\n### Naming best practices\n\n- Use the folder to reflect ownership, not to repeat the item type. `finance\u002Fuse-net-revenue` is good. `instructions\u002Fuse-net-revenue` is redundant.\n- Name after the concept or rule, not the implementation. `exclude-canceled-orders` is better than `filter-status-not-canceled`.\n- For memory events, name after the event itself, not a generic label. `revenue-redefinition` or `gdpr-rollout` beats `data-change-3`.\n\n---\n\n## Frontmatter format\n\nContext items are stored as frontmatter documents: a YAML block between `---` delimiters, followed by a markdown prose body. When calling `create_context_item`, pass the YAML fields as `context_item` parameters and the prose body as the `markdown_text` parameter.\n\nThe `description` field is the **retrieval signal** — the model reads it to decide whether to fetch an on-demand item. It is **used for skills, knowledge, and memory events**. Do not add `description` to instructions; their only content is the prose body.\n\nSee [examples.md](examples.md) for what the full frontmatter documents look like.\n\n---\n\n## Type Reference\n\n### Instruction — standing rule (always on)\n\n**Use when:** There is a standing preference, constraint, or fact that the model must respect in every conversation. These are injected into every prompt automatically.\n\n**Keep instructions concise.** They are never filtered out, so verbose instructions bloat every prompt. If the content is more than a short paragraph, turn it into a skill instead.\n\n**Parameters:**\n- `type`: `instruction`\n- `subtype`: `instruction`\n- `name`: `folder\u002Frule-name`\n- `title`: short human label\n- prose body: **the rule text** — what the model must do or avoid\n- `related_objects`: *(optional)* only the fields or entities directly named or constrained by this instruction — not downstream consumers, not lineage, not entities that happen to join to a relevant one; use an entity reference instead of a field list when the instruction applies to most fields of that entity\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller; only set if a different team or person should be listed\n\n**Good instruction topics:**\n- Metric selection preferences (\"always use net_revenue\")\n- Domain-specific constraints (\"exclude test accounts from all analyses\")\n- Terminology corrections (\"what users call 'bookings' maps to the `orders.gmv` metric\")\n\n### Skill — on-demand playbook\n\n**Use when:** There is a multi-step analysis procedure the model should follow when asked about a specific topic. Skills are retrieved when the model judges them relevant based on the description.\n\n**Write a rich, keyword-dense `description`** — this is what the model reads to decide whether to retrieve the skill. A weak description means the skill will be missed.\n\n**Parameters:**\n- `type`: `instruction`\n- `subtype`: `skill`\n- `name`: `topic-or-folder\u002Ftopic`\n- `title`: short human label\n- `description`: **required** — precise, keyword-rich description of when to use this skill and what it covers (this is the retrieval signal)\n- prose body: **the playbook** — step-by-step guide written in markdown (goes after the closing `---`)\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller\n\n**Good skill topics:**\n- Root cause analysis procedures\n- Customer cohort or churn investigation workflows\n- Reporting preparation checklists\n- Domain-specific investigation playbooks\n\n### Knowledge — external source pointer\n\n**Use when:** There is an external document (Confluence page, Notion database, runbook, policy) that the model should consult when answering relevant questions. The model fetches the content via an MCP server at retrieval time.\n\n**Write a precise `description`** that includes the topics the document covers — this is the retrieval signal.\n\n**Parameters:**\n- `type`: `instruction`\n- `subtype`: `knowledge`\n- `name`: `source-folder\u002Fdocument-name`\n- `title`: short human label\n- `description`: **required** — what topics this document covers (used for retrieval)\n- `external_source.tool`: the MCP server name (e.g. `confluence`, `notion`)\n- `external_source.resource_id`: the URI within that server (e.g. `confluence:\u002F\u002FPage-Title`)\n- prose body: typically omit — the content is fetched from the external source at runtime\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller\n\n### Memory \u002F Event — historical record\n\n**Use when:** Something happened that affects how data should be interpreted: a metric was redefined, a process changed, a data migration occurred, a business event shifted baseline numbers.\n\nMemory items are retrieved on demand when the model judges them relevant to the current question.\n\n**Parameters:**\n- `type`: `memory`\n- `subtype`: `event`\n- `name`: `domain\u002Fevent-name`\n- `title`: short human label\n- `description`: **required** — what this event is about (used for retrieval)\n- prose body: **what happened and why it matters** for data interpretation\n- `from_date`: when the event occurred (point-in-time) or when it started (if a duration)\n- `to_date`: *(optional)* end date for events spanning a period\n- `related_objects`: *(optional)* only the fields or entities directly changed or affected by this event — not downstream consumers, not lineage, not entities that happen to join to a relevant one; use an entity reference instead of a field list when the event affects most fields of that entity\n- `labels`: optional tags for grouping\n- `owner`: *(optional)* defaults to the caller\n\n**Good memory topics:**\n- Metric or definition changes with a specific effective date\n- ETL migrations or data backfills\n- Significant business events (acquisitions, product launches) that cause discontinuities in data\n- Changes in how data is collected or tagged\n\n---\n\n## Examples\n\nSee [examples.md](examples.md) for full worked examples of each type, plus right-vs-wrong comparisons.\n\n---\n\n## Discovery Helpers\n\n- `list_context_items` — see all existing items (avoid duplicates)\n- `get_context_item` — read a specific item's current content before updating\n- `list_agents` \u002F `get_agent` — see which agents reference which context globs, to understand the scoping impact of adding a new item\n\n---\n\n## Documentation Lookup\n\nUse the `honeydew-docs` MCP tools to search the Honeydew documentation when:\n\n- You need to understand how context items interact with agents or how retrieval works\n- The user asks about agent configuration, glob patterns, or context scoping\n- You need guidance on memory subtypes or advanced context item options\n\nSearch for topics like: \"context items\", \"instructions\", \"agent context\", \"memory events\", \"skills\".\n\n---\n\n## Best Practices\n\n1. **Semantic layer first.** If the request is about logic, calculations, definitions, or data structure — build it in the semantic layer. Context items reference semantic objects; they do not replace them.\n\n2. **Instructions must be short.** They're injected into every prompt. If it doesn't fit in 2–3 sentences, turn it into a skill.\n\n3. **Skill descriptions are the retrieval key.** Write rich, specific, keyword-heavy descriptions. \"Guide for root cause analysis of revenue drops, covering cohort isolation, period-over-period comparison, and funnel decomposition\" is far better than \"revenue analysis\".\n\n4. **Folder by domain, not by type.** `finance\u002F` should contain instructions, skills, knowledge, and events all related to finance. This makes agent glob patterns meaningful.\n\n5. **Check for duplicates before creating.** Run `list_context_items` first. Overlapping or contradictory instructions are worse than none.\n\n6. **Use labels for cross-cutting concerns.** Labels like `pii`, `finance`, or `deprecated` enable filtering that cuts across folder hierarchy.\n\n---\n\n## MANDATORY: Confirm You Are Not Duplicating Semantic Layer Logic\n\nBefore creating any context item, verify:\n\n1. **Does this describe a calculation or data definition?** → Build it as a metric, attribute, or filter instead.\n2. **Does this duplicate something already in the semantic layer?** → Update the semantic object's `description` field instead of adding context.\n3. **Is this a preference that references an existing metric by name?** → Good candidate for an instruction.\n\nIf you are unsure, ask the user: *\"Should I add this as a rule for the AI analyst, or should I encode it directly in the semantic layer as a metric\u002Fattribute?\"*\n",{"name":5,"description":7},"main","2b63f856ce19c2a797929cacabd9e0fc936812fa","plugins\u002Fhoneydew-ai\u002Fhoneydew-ai-coding-agents-plugins\u002Fskills\u002Fcontext-item-creation",491,"https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins","https:\u002F\u002Fhol.org\u002Fplugins\u002Fbest-codex-plugins","https:\u002F\u002Fgithub.com\u002Fhashgraph-online","https:\u002F\u002Fhol.org","Open-source standards and developer tools for interoperable AI agents: discovery, identity, communication, payments, privacy and security.","2026-11-14 14:06:13",{"source":8,"skillName":5,"commands":61,"prompt":71,"zipAvailable":72},[62,65,68],{"tool":63,"command":64},"npx","npx skills add hashgraph-online\u002Fawesome-codex-plugins --skill context-item-creation",{"tool":66,"command":67},"bunx","bunx skills add hashgraph-online\u002Fawesome-codex-plugins --skill context-item-creation",{"tool":69,"command":70},"pnpm","pnpm dlx skills add hashgraph-online\u002Fawesome-codex-plugins --skill context-item-creation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fhoneydew-ai\u002Fhoneydew-ai-coding-agents-plugins\u002Fskills\u002Fcontext-item-creation\u002FSKILL.md",true,[74],{"path":75,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":78,"sourceHtmlUrl":79},"examples.md","REFERENCE","参考文档",8978,"https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fhoneydew-ai\u002Fhoneydew-ai-coding-agents-plugins\u002Fskills\u002Fcontext-item-creation\u002Fexamples.md",[81,91,99,107,115,123],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":85,"licenseFileUrl":86,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":90},"01M7DV332YQMGD8WTRJ9DZQXTB","worktable-review-with-annotations","Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.","MIT","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-review-with-annotations\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-review-with-annotations\u002FSKILL.md",[],[],"2026-11-15 14:19:17",{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":85,"licenseFileUrl":95,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":90},"01M7DV2YGHCZVBEB1AHYB6CGHE","worktable-find-and-synthesize-context","Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-find-and-synthesize-context\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-find-and-synthesize-context\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":85,"licenseFileUrl":103,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":90},"01M7DV2TQN1H11ME2SAF9MXN7V","worktable-create-or-update-docs","Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-update-docs\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-update-docs\u002FSKILL.md",[],[],{"skillNo":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":85,"licenseFileUrl":111,"sourceHtmlUrl":112,"categories":113,"keywords":114,"publishedAt":90},"01M7DV2NE452NEA8TXATKSNAJH","worktable-create-or-manage-records","Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-records\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-records\u002FSKILL.md",[],[],{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":85,"licenseFileUrl":119,"sourceHtmlUrl":120,"categories":121,"keywords":122,"publishedAt":90},"01M7DV2HENRBR2KSCVM064WH0Y","worktable-create-or-manage-interactive-html","Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-interactive-html\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-create-or-manage-interactive-html\u002FSKILL.md",[],[],{"skillNo":124,"name":125,"nameZh":6,"description":126,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":85,"licenseFileUrl":127,"sourceHtmlUrl":128,"categories":129,"keywords":130,"publishedAt":90},"01M7DV2CADR5R7RQFFYPNAGR9Y","worktable-collaborate-threads","Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-collaborate-threads\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fhashgraph-online\u002Fawesome-codex-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fworktable\u002Fworktable-plugin\u002Fskills\u002Fworktable-collaborate-threads\u002FSKILL.md",[],[]]