[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74MFV4QP93ZAA5MJMV28VWK":3,"skill-same:01M74MFV4QP93ZAA5MJMV28VWK":47},{"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":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M74MFV4QP93ZAA5MJMV28VWK","self-awareness",null,"Claude Science's own session database schema and SDK surface for introspection via host.query(). Load this when you need to query your own conversation history, token usage, cost accounting, execution log, or artifact metadata beyond what host.frames()\u002Fhost.artifacts() provide — e.g. \"how many tokens has this session used\", \"what was my last tool call\", \"list every file I've written\", \"where are messages stored\", \"what tables can I query\", \"inspect frames.context_data\", or any time you're about to PRAGMA-probe the Claude Science metadata DB to discover its schema.","pengqianhan\u002FAI-Human-Research-OS","pengqianhan","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F4286531?v=4","01M74M3DT9HNZ6HPAK8TX69EX9","PengqianHan","01M74M3DTVSN2GXW0JH2W396FK",3,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Fself-awareness\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: self-awareness\ndescription: Claude Science's own session database schema and SDK surface for introspection via host.query(). Load this when you need to query your own conversation history, token usage, cost accounting, execution log, or artifact metadata beyond what host.frames()\u002Fhost.artifacts() provide — e.g. \"how many tokens has this session used\", \"what was my last tool call\", \"list every file I've written\", \"where are messages stored\", \"what tables can I query\", \"inspect frames.context_data\", or any time you're about to PRAGMA-probe the Claude Science metadata DB to discover its schema.\nlicense: Apache-2.0\n---\n\n# Self-awareness — Claude Science's own database and SDK\n\n`host.query(sql, params=[], limit=None, df=False)` runs read-only SQLite\nagainst Claude Science's own metadata DB. It is only available via the **`repl`\ntool** (not `python`\u002F`r`). Results are automatically scoped to the current\nproject, so `SELECT * FROM frames` returns only frames in this project. The\n`repl` tool is stdlib-only — `df=True` returns the raw dict there (use\n`json.dump(..., open(\"handoff\u002Fq.json\",\"w\"))` and load in a `python` cell if\nyou want pandas).\n\n## Dialect and limits\n\n- **SQLite.** Epoch-milliseconds for all timestamps\n  (`created_at > strftime('%s','now','-1 day')*1000`). Booleans are `0`\u002F`1`.\n  JSON columns are TEXT — use `json_extract(col, '$.key')`. Recursive CTEs OK.\n- `SELECT` \u002F `WITH` \u002F `PRAGMA` \u002F `EXPLAIN` only; one statement per call;\n  `?` placeholders with `params=[...]`.\n- **Scoping.** Most tables are transparently filtered to the current\n  project (and `memories` to the current user) via CTEs that shadow the\n  real tables — `session_claims`, `verification_checks`, and `poller_lease`\n  are unscoped. You therefore **cannot** use `main.table` \u002F `temp.table` —\n  schema-qualified names are rejected.\n- **Caps.** Default 200 rows (max `limit=1000`); cells >2000 chars are\n  clipped in place with a `…[+N chars]` marker; total serialized output\n  capped at ~100k chars (`truncated=True`, `truncation_reason=\"total_size_cap\"`\n  — narrow your columns). 5-second timeout.\n- Schema introspection: `host.query(\"PRAGMA table_info(frames)\")` or\n  `host.query(\"SELECT name, sql FROM sqlite_master WHERE type='table'\")`.\n\n## Queryable tables\n\n### Session \u002F conversation\n\n**`frames`** — one row per agent frame (a root conversation or a delegated\nsub-agent). The frame you are running in now is one of these rows.\nKey columns: `id`, `parent_frame_id`, `root_frame_id`, `agent_name`,\n`delegate_name`, `status` (`processing`\u002F`completed`\u002F`failed`\u002F`cancelled`\u002F\n`awaiting_user_response`\u002F`awaiting_plan_approval`), `model`, `effort`, `input_tokens`,\n`output_tokens`, `cache_read_tokens`, `cache_write_tokens`, `total_cost`,\n`task_summary`, `status_description`, `conversation_type`, `name`,\n`project_id`, `created_at`, `updated_at`, `completed_at`,\n`last_user_message_at`, `is_hidden`.\nJSON columns: `input_data` (what started the frame), `output_data`\n(`json_extract(output_data,'$.response')` is the final response text),\n`context_data` (the full serialized runner state — see below),\n`mentioned_artifact_ids`, `specialists_used`.\n\n`context_data` is large. It holds the entire runner state under\nunderscore-prefixed keys — notably `$._messages` (the full conversation\narray), `$._input_tokens` \u002F `$._output_tokens` \u002F `$._total_cost` (same values\nas the top-level columns), `$._running_children`, `$._plan_json`,\n`$._compaction_count`, `$._tool_id_to_frame_id`. Selecting it raw will hit\nthe cell cap; use `json_extract`\u002F`json_array_length` to read specific keys.\nFor the messages themselves, prefer `host.frames(frame_id=...)` which\npaginates — `_messages` via SQL will truncate on any non-trivial session.\n\n**`compaction_archives`** — pre-compaction message snapshots.\n`frame_id`, `compaction_index`, `message_count`, `token_count`, `summary`,\n`messages` (JSON array), `created_at`. When a frame's `_compaction_count > 0`,\nthe original messages that were summarized live here.\n\n**`notifications`** — parent↔child messages. `sender_frame_id`,\n`recipient_frame_id`, `root_frame_id`, `notification_type`, `payload` (JSON),\n`read_at`, `created_at`.\n\n**`projects`** — `id` (`proj_*`, not a UUID), `name`, `description`,\n`context`, `user_id`, `uploads_frame_id`, `memory_enabled`, `created_at`,\n`updated_at`.\n\n**`notes`** — user annotations. `project_id`, `target_type`,\n`target_frame_id`, `target_message_index`, `target_artifact_id`, `content`.\n\n### Artifacts\n\n**`artifacts`** — one row per file. `id`, `project_id`, `root_frame_id`,\n`frame_id`, `filename`, `latest_version_id`, `is_user_upload`, `is_ephemeral`,\n`folder_id`, `sort_order`, `priority`, `created_at`.\n\n**`artifact_versions`** — one row per saved revision. `id`, `artifact_id`,\n`version_number`, `frame_id`, `content_type`, `size_bytes`, `checksum`,\n`storage_path`, `extracted_code`, `code_description`, `language`,\n`agent_name`, `is_intermediate`, `is_checkpoint`, `parent_version_id`,\n`producing_cell_id` (→ `execution_log.id`), `created_at`. JSON:\n`lineage_messages`, `dependency_mappings`, `environment_snapshot`,\n`annotations`, `cell_sources`. Join `artifacts.latest_version_id =\nartifact_versions.id` for size\u002Ftype.\n\n**`artifact_dependencies`** — DAG edges. `artifact_version_id`,\n`depends_on_version_id`, `reference_name`.\n\n**`artifact_folders`** — `id`, `project_id`, `parent_id`, `name`,\n`root_frame_id`, `is_conversation_folder`, `is_user_uploads_folder`,\n`sort_order`.\n\n**`content_snapshots`** — content-addressed dedup store. `hash`, `content`,\n`size_bytes`. Referenced by `artifact_versions.lineage_snapshot_hash` \u002F\n`env_snapshot_hash`.\n\n### Execution history\n\n**`execution_log`** — one row per `python`\u002F`r`\u002F`bash`\u002F`repl` cell, in\norder. `id`, `frame_id`, `cell_index` (monotonic), `kernel_id`, `kernel_kind`\n(`analysis`\u002F`operon`), `conda_env`,\n`language`, `source` (exact submitted\ncode), `stdout`, `stderr`, `exit_status` (`ok`\u002F`error`\u002F`kernel_died`\u002F\n`cancelled`), `error_lineno`, `files_written` (JSON `[{path, sha256}]`),\n`created_at`. This is the ground-truth record of everything you've run.\n\n**`host_call_log`** — one row per `host.*` SDK call made inside a cell.\n`id`, `execution_log_id` (→ `execution_log.id`), `seq`, `method`\n(`query_db`\u002F`llm`\u002F`mcp`\u002F`list_frames`\u002F…), `args_json`, `derivable`,\n`data_inline`, `data_ref`, `error`, `bytes`, `created_at`. Ordered by\n`(execution_log_id, seq)`.\n\n### Compute and verification\n\n**`compute_usage`** — remote compute jobs. `job_id`, `environment`,\n`tier_type` (`gpu`\u002F`cpu`), `provider`, `frame_id`, `project_id`, `started_at`,\n`ended_at` (null ⇒ running), `expires_at`, `state`, `remote_workdir`,\n`submit_cell_id`. JSON: `output_specs`, `remote_handle`.\n\n**`session_claims`** — falsifiable claims extracted for verification.\n`root_frame_id`, `frame_id`, `step_id`, `claim_text`, `entities` (JSON),\n`source` (`agent`\u002F`haiku_extracted`).\n\n**`verification_checks`** — reviewer verdicts. `root_frame_id`,\n`artifact_version_id`, `claim_id`, `claim`, `verdict`\n(`pass`\u002F`warn`\u002F`fail`\u002F`inconclusive`), `severity`, `evidence`, `rebuttal`,\n`reviewer_model`, `reviewer_frame_id`, `source_ref` (JSON), `status`\n(`open`\u002F`resolved`\u002F`unaddressed`), `reflag_count`.\n\n**`memories`** — durable beliefs (user-scoped; may be absent on some builds).\n`id` (`mem_*`), `body`, `subject_project_id`, `subject_artifact_id`,\n`subject_version_id`, `subject_frame_id`, `source_frame_id`, `origin`\n(`extractor`\u002F`agent_tool`\u002F`user`), `evidence`\n(`stated`\u002F`observed`\u002F`inferred`), `superseded_by`, `last_surfaced_at`.\n\n**`poller_lease`** — single-writer guard for compute polling. `provider`,\n`holder`, `expires_at`.\n\n## Denied tables\n\nThese are rejected with `Table '\u003Cname>' is not queryable` — use the listed\nSDK accessor instead.\n\n- Secrets (encrypted at rest, blocked defense-in-depth): `oauth_tokens`,\n  `user_secrets`, `anthropic_api_keys`, `cloud_credentials`. →\n  `host.credentials.list()` for non-secret metadata; `.get(name)` for\n  the decrypted fields — usable in client libraries, redacted only from\n  printed cell output.\n- Agent\u002Fskill\u002Fconnector configuration (enumerating attack surface has no\n  legitimate raw-SQL use): `user_agents`, `agents`, `custom_agent_prompts`,\n  `bundled_agent_settings`, `capability_settings`, `custom_skills`,\n  `agent_skill_assignments`, `custom_mcp_servers`, `mcp_agent_assignments`,\n  `mcp_tool_grants`, `directory_attachments`. → `host.agents.list()` \u002F\n  `host.skills.list()` \u002F `host.agents.list_connectors()` (load the\n  `customize` skill for that API).\n- Host filesystem mounts: `host_grants`. → the `list_host_grants` tool\n  (present on sandboxed-network builds).\n- Compute provider configuration: `compute_providers`. → the\n  `list_compute` \u002F `compute_details` tools.\n\nThe denylist matches on word boundaries anywhere in the SQL, so a column\nalias or string literal that happens to equal a denied table name will also\nbe rejected.\n\nHost identity (hostname, workspace\u002Fpod name) is intentionally not exposed\nanywhere in this DB (and on Linux builds the sandbox masks it as well) — to\nknow where you're running, ask the user or use `list_compute` labels.\n\n## Worked examples\n\nAll of these run via the `repl` tool.\n\n```python\n# Token and cost accounting across every frame in THIS PROJECT (all\n# sessions). Add `WHERE root_frame_id = ?` with the current root's id to\n# scope to one session tree. Aggregate server-side so the row cap can't\n# undercount.\nr = host.query(\"\"\"\n  SELECT COUNT(*)                   AS n_frames,\n         SUM(input_tokens)          AS input_tokens,\n         SUM(output_tokens)         AS output_tokens,\n         SUM(cache_read_tokens)     AS cache_read_tokens,\n         SUM(cache_write_tokens)    AS cache_write_tokens,\n         SUM(total_cost)            AS total_cost\n  FROM frames\n\"\"\")\nn, itok, otok, crd, cwr, cost = r[\"rows\"][0]\nprint(f\"{n} frames, ${cost or 0:.4f} total\")\n```\n\n```python\n# Last 10 code cells executed in this project (any frame), with outcome.\n# Add `WHERE e.frame_id = ?` with the current frame's id to scope to one\n# frame.\nhost.query(\"\"\"\n  SELECT e.frame_id, e.cell_index, e.language, e.kernel_kind, e.conda_env,\n         e.exit_status, substr(e.source, 1, 120) AS src,\n         json_array_length(e.files_written) AS n_files\n  FROM execution_log e\n  ORDER BY e.created_at DESC\n  LIMIT 10\n\"\"\")\n```\n\n```python\n# How far into context is each root conversation in this project? Reads\n# _messages length and compaction count without pulling the whole blob.\nhost.query(\"\"\"\n  SELECT id, name,\n         json_array_length(context_data, '$._messages')   AS n_messages,\n         json_extract(context_data, '$._compaction_count') AS compactions,\n         input_tokens, output_tokens\n  FROM frames\n  WHERE parent_frame_id IS NULL\n  ORDER BY updated_at DESC\n\"\"\")\n```\n\n```python\n# Every artifact this project has, with current size\u002Ftype, newest first.\nhost.query(\"\"\"\n  SELECT a.filename, v.content_type, v.size_bytes, v.version_number,\n         a.is_user_upload, a.latest_version_id\n  FROM artifacts a\n  JOIN artifact_versions v ON a.latest_version_id = v.id\n  WHERE a.is_ephemeral = 0\n  ORDER BY v.created_at DESC\n\"\"\")\n```\n\n## SDK surface — which tool runs what\n\nThe `host` object is a Python SDK backed by host-side RPCs. Run\n`help(host)` \u002F `help(host.\u003Cx>)` for signatures.\n\n| Accessor | Tool | Returns |\n|---|---|---|\n| `host.query(sql, params, limit, df)` | **`repl`** | Raw SQL over the tables above |\n| `host.frames(...)` | **`repl`** | List\u002Fsearch\u002Fdetail frames (paginated messages) |\n| `host.children()` | **`repl`** | Live sub-agents (delegation-enabled profiles only) |\n| `host.delegate(task_or_list, name=?, profile=?, output_schema=?, model=?)` | **`repl`** | Spawn child agent(s), block until done (ultra-mode roots; requires `[delegation] sdk_enabled`). `model=` pins the child's model per request — e.g. a haiku-class id for cheap fan-outs. Blocks the cell — for long-running children run it in a background cell (a user message mid-call backgrounds it; a Stop \u002F cell interrupt cancels the children) |\n| `host.agents.*` \u002F `host.skills.*` | **`repl`** | Profile and skill CRUD — load `customize` skill |\n| `host.submit_output(output, completion_bullets=[...])` | **`repl`** | Submit your structured result when your task carries an OUTPUT SCHEMA section (required before completing). Build the dict in-kernel — the payload rides the host-call wire, not your prose; on a validation\u002Freview bounce, mutate the dict in memory and resubmit (replaces the recorded output) |\n| `host.compute.*` | **`repl`** | Remote job submit\u002Fwait — load the compute skill it names |\n| `host.artifacts(...)` | `python` | Filtered artifact search (wraps the join above) |\n| `host.artifact_path(vid)` \u002F `host.artifact_marker(vid)` | `python` | Resolve a version_id to a readable path \u002F marker |\n| `host.lineage[vid]` | `python` | `{code, messages, env, inputs}` for one version |\n| `host.llm(prompt_or_list, model=?, ...)` | `python` | Single-turn completion via the host's API client. Omitting `model=` uses the Haiku-class kernel default (via `[llm] kernel_default_model`); for harder reasoning pass `model=host.current_model()` — never hardcode a literal model id (they go stale) |\n| `host.credentials.list()` \u002F `.get(name)` | `python` | User-configured credential metadata |\n| `host.mcp(server, method, **kw)` | **`repl`** | MCP\u002Fconnector call — only exists in the `repl` tool; pass results to `python`\u002F`r` via `.\u002Fhandoff\u002F*.json` |\n\nThe `repl` tool and the `python` tool are separate processes that share\nonly the workspace directory — move data between them via\n`.\u002Fhandoff\u002F*.json`, not variables.\n","# Self-awareness — Claude Science's own database and SDK\n\n`host.query(sql, params=[], limit=None, df=False)` runs read-only SQLite\nagainst Claude Science's own metadata DB. It is only available via the **`repl`\ntool** (not `python`\u002F`r`). Results are automatically scoped to the current\nproject, so `SELECT * FROM frames` returns only frames in this project. The\n`repl` tool is stdlib-only — `df=True` returns the raw dict there (use\n`json.dump(..., open(\"handoff\u002Fq.json\",\"w\"))` and load in a `python` cell if\nyou want pandas).\n\n## Dialect and limits\n\n- **SQLite.** Epoch-milliseconds for all timestamps\n  (`created_at > strftime('%s','now','-1 day')*1000`). Booleans are `0`\u002F`1`.\n  JSON columns are TEXT — use `json_extract(col, '$.key')`. Recursive CTEs OK.\n- `SELECT` \u002F `WITH` \u002F `PRAGMA` \u002F `EXPLAIN` only; one statement per call;\n  `?` placeholders with `params=[...]`.\n- **Scoping.** Most tables are transparently filtered to the current\n  project (and `memories` to the current user) via CTEs that shadow the\n  real tables — `session_claims`, `verification_checks`, and `poller_lease`\n  are unscoped. You therefore **cannot** use `main.table` \u002F `temp.table` —\n  schema-qualified names are rejected.\n- **Caps.** Default 200 rows (max `limit=1000`); cells >2000 chars are\n  clipped in place with a `…[+N chars]` marker; total serialized output\n  capped at ~100k chars (`truncated=True`, `truncation_reason=\"total_size_cap\"`\n  — narrow your columns). 5-second timeout.\n- Schema introspection: `host.query(\"PRAGMA table_info(frames)\")` or\n  `host.query(\"SELECT name, sql FROM sqlite_master WHERE type='table'\")`.\n\n## Queryable tables\n\n### Session \u002F conversation\n\n**`frames`** — one row per agent frame (a root conversation or a delegated\nsub-agent). The frame you are running in now is one of these rows.\nKey columns: `id`, `parent_frame_id`, `root_frame_id`, `agent_name`,\n`delegate_name`, `status` (`processing`\u002F`completed`\u002F`failed`\u002F`cancelled`\u002F\n`awaiting_user_response`\u002F`awaiting_plan_approval`), `model`, `effort`, `input_tokens`,\n`output_tokens`, `cache_read_tokens`, `cache_write_tokens`, `total_cost`,\n`task_summary`, `status_description`, `conversation_type`, `name`,\n`project_id`, `created_at`, `updated_at`, `completed_at`,\n`last_user_message_at`, `is_hidden`.\nJSON columns: `input_data` (what started the frame), `output_data`\n(`json_extract(output_data,'$.response')` is the final response text),\n`context_data` (the full serialized runner state — see below),\n`mentioned_artifact_ids`, `specialists_used`.\n\n`context_data` is large. It holds the entire runner state under\nunderscore-prefixed keys — notably `$._messages` (the full conversation\narray), `$._input_tokens` \u002F `$._output_tokens` \u002F `$._total_cost` (same values\nas the top-level columns), `$._running_children`, `$._plan_json`,\n`$._compaction_count`, `$._tool_id_to_frame_id`. Selecting it raw will hit\nthe cell cap; use `json_extract`\u002F`json_array_length` to read specific keys.\nFor the messages themselves, prefer `host.frames(frame_id=...)` which\npaginates — `_messages` via SQL will truncate on any non-trivial session.\n\n**`compaction_archives`** — pre-compaction message snapshots.\n`frame_id`, `compaction_index`, `message_count`, `token_count`, `summary`,\n`messages` (JSON array), `created_at`. When a frame's `_compaction_count > 0`,\nthe original messages that were summarized live here.\n\n**`notifications`** — parent↔child messages. `sender_frame_id`,\n`recipient_frame_id`, `root_frame_id`, `notification_type`, `payload` (JSON),\n`read_at`, `created_at`.\n\n**`projects`** — `id` (`proj_*`, not a UUID), `name`, `description`,\n`context`, `user_id`, `uploads_frame_id`, `memory_enabled`, `created_at`,\n`updated_at`.\n\n**`notes`** — user annotations. `project_id`, `target_type`,\n`target_frame_id`, `target_message_index`, `target_artifact_id`, `content`.\n\n### Artifacts\n\n**`artifacts`** — one row per file. `id`, `project_id`, `root_frame_id`,\n`frame_id`, `filename`, `latest_version_id`, `is_user_upload`, `is_ephemeral`,\n`folder_id`, `sort_order`, `priority`, `created_at`.\n\n**`artifact_versions`** — one row per saved revision. `id`, `artifact_id`,\n`version_number`, `frame_id`, `content_type`, `size_bytes`, `checksum`,\n`storage_path`, `extracted_code`, `code_description`, `language`,\n`agent_name`, `is_intermediate`, `is_checkpoint`, `parent_version_id`,\n`producing_cell_id` (→ `execution_log.id`), `created_at`. JSON:\n`lineage_messages`, `dependency_mappings`, `environment_snapshot`,\n`annotations`, `cell_sources`. Join `artifacts.latest_version_id =\nartifact_versions.id` for size\u002Ftype.\n\n**`artifact_dependencies`** — DAG edges. `artifact_version_id`,\n`depends_on_version_id`, `reference_name`.\n\n**`artifact_folders`** — `id`, `project_id`, `parent_id`, `name`,\n`root_frame_id`, `is_conversation_folder`, `is_user_uploads_folder`,\n`sort_order`.\n\n**`content_snapshots`** — content-addressed dedup store. `hash`, `content`,\n`size_bytes`. Referenced by `artifact_versions.lineage_snapshot_hash` \u002F\n`env_snapshot_hash`.\n\n### Execution history\n\n**`execution_log`** — one row per `python`\u002F`r`\u002F`bash`\u002F`repl` cell, in\norder. `id`, `frame_id`, `cell_index` (monotonic), `kernel_id`, `kernel_kind`\n(`analysis`\u002F`operon`), `conda_env`,\n`language`, `source` (exact submitted\ncode), `stdout`, `stderr`, `exit_status` (`ok`\u002F`error`\u002F`kernel_died`\u002F\n`cancelled`), `error_lineno`, `files_written` (JSON `[{path, sha256}]`),\n`created_at`. This is the ground-truth record of everything you've run.\n\n**`host_call_log`** — one row per `host.*` SDK call made inside a cell.\n`id`, `execution_log_id` (→ `execution_log.id`), `seq`, `method`\n(`query_db`\u002F`llm`\u002F`mcp`\u002F`list_frames`\u002F…), `args_json`, `derivable`,\n`data_inline`, `data_ref`, `error`, `bytes`, `created_at`. Ordered by\n`(execution_log_id, seq)`.\n\n### Compute and verification\n\n**`compute_usage`** — remote compute jobs. `job_id`, `environment`,\n`tier_type` (`gpu`\u002F`cpu`), `provider`, `frame_id`, `project_id`, `started_at`,\n`ended_at` (null ⇒ running), `expires_at`, `state`, `remote_workdir`,\n`submit_cell_id`. JSON: `output_specs`, `remote_handle`.\n\n**`session_claims`** — falsifiable claims extracted for verification.\n`root_frame_id`, `frame_id`, `step_id`, `claim_text`, `entities` (JSON),\n`source` (`agent`\u002F`haiku_extracted`).\n\n**`verification_checks`** — reviewer verdicts. `root_frame_id`,\n`artifact_version_id`, `claim_id`, `claim`, `verdict`\n(`pass`\u002F`warn`\u002F`fail`\u002F`inconclusive`), `severity`, `evidence`, `rebuttal`,\n`reviewer_model`, `reviewer_frame_id`, `source_ref` (JSON), `status`\n(`open`\u002F`resolved`\u002F`unaddressed`), `reflag_count`.\n\n**`memories`** — durable beliefs (user-scoped; may be absent on some builds).\n`id` (`mem_*`), `body`, `subject_project_id`, `subject_artifact_id`,\n`subject_version_id`, `subject_frame_id`, `source_frame_id`, `origin`\n(`extractor`\u002F`agent_tool`\u002F`user`), `evidence`\n(`stated`\u002F`observed`\u002F`inferred`), `superseded_by`, `last_surfaced_at`.\n\n**`poller_lease`** — single-writer guard for compute polling. `provider`,\n`holder`, `expires_at`.\n\n## Denied tables\n\nThese are rejected with `Table '\u003Cname>' is not queryable` — use the listed\nSDK accessor instead.\n\n- Secrets (encrypted at rest, blocked defense-in-depth): `oauth_tokens`,\n  `user_secrets`, `anthropic_api_keys`, `cloud_credentials`. →\n  `host.credentials.list()` for non-secret metadata; `.get(name)` for\n  the decrypted fields — usable in client libraries, redacted only from\n  printed cell output.\n- Agent\u002Fskill\u002Fconnector configuration (enumerating attack surface has no\n  legitimate raw-SQL use): `user_agents`, `agents`, `custom_agent_prompts`,\n  `bundled_agent_settings`, `capability_settings`, `custom_skills`,\n  `agent_skill_assignments`, `custom_mcp_servers`, `mcp_agent_assignments`,\n  `mcp_tool_grants`, `directory_attachments`. → `host.agents.list()` \u002F\n  `host.skills.list()` \u002F `host.agents.list_connectors()` (load the\n  `customize` skill for that API).\n- Host filesystem mounts: `host_grants`. → the `list_host_grants` tool\n  (present on sandboxed-network builds).\n- Compute provider configuration: `compute_providers`. → the\n  `list_compute` \u002F `compute_details` tools.\n\nThe denylist matches on word boundaries anywhere in the SQL, so a column\nalias or string literal that happens to equal a denied table name will also\nbe rejected.\n\nHost identity (hostname, workspace\u002Fpod name) is intentionally not exposed\nanywhere in this DB (and on Linux builds the sandbox masks it as well) — to\nknow where you're running, ask the user or use `list_compute` labels.\n\n## Worked examples\n\nAll of these run via the `repl` tool.\n\n```python\n# Token and cost accounting across every frame in THIS PROJECT (all\n# sessions). Add `WHERE root_frame_id = ?` with the current root's id to\n# scope to one session tree. Aggregate server-side so the row cap can't\n# undercount.\nr = host.query(\"\"\"\n  SELECT COUNT(*)                   AS n_frames,\n         SUM(input_tokens)          AS input_tokens,\n         SUM(output_tokens)         AS output_tokens,\n         SUM(cache_read_tokens)     AS cache_read_tokens,\n         SUM(cache_write_tokens)    AS cache_write_tokens,\n         SUM(total_cost)            AS total_cost\n  FROM frames\n\"\"\")\nn, itok, otok, crd, cwr, cost = r[\"rows\"][0]\nprint(f\"{n} frames, ${cost or 0:.4f} total\")\n```\n\n```python\n# Last 10 code cells executed in this project (any frame), with outcome.\n# Add `WHERE e.frame_id = ?` with the current frame's id to scope to one\n# frame.\nhost.query(\"\"\"\n  SELECT e.frame_id, e.cell_index, e.language, e.kernel_kind, e.conda_env,\n         e.exit_status, substr(e.source, 1, 120) AS src,\n         json_array_length(e.files_written) AS n_files\n  FROM execution_log e\n  ORDER BY e.created_at DESC\n  LIMIT 10\n\"\"\")\n```\n\n```python\n# How far into context is each root conversation in this project? Reads\n# _messages length and compaction count without pulling the whole blob.\nhost.query(\"\"\"\n  SELECT id, name,\n         json_array_length(context_data, '$._messages')   AS n_messages,\n         json_extract(context_data, '$._compaction_count') AS compactions,\n         input_tokens, output_tokens\n  FROM frames\n  WHERE parent_frame_id IS NULL\n  ORDER BY updated_at DESC\n\"\"\")\n```\n\n```python\n# Every artifact this project has, with current size\u002Ftype, newest first.\nhost.query(\"\"\"\n  SELECT a.filename, v.content_type, v.size_bytes, v.version_number,\n         a.is_user_upload, a.latest_version_id\n  FROM artifacts a\n  JOIN artifact_versions v ON a.latest_version_id = v.id\n  WHERE a.is_ephemeral = 0\n  ORDER BY v.created_at DESC\n\"\"\")\n```\n\n## SDK surface — which tool runs what\n\nThe `host` object is a Python SDK backed by host-side RPCs. Run\n`help(host)` \u002F `help(host.\u003Cx>)` for signatures.\n\n| Accessor | Tool | Returns |\n|---|---|---|\n| `host.query(sql, params, limit, df)` | **`repl`** | Raw SQL over the tables above |\n| `host.frames(...)` | **`repl`** | List\u002Fsearch\u002Fdetail frames (paginated messages) |\n| `host.children()` | **`repl`** | Live sub-agents (delegation-enabled profiles only) |\n| `host.delegate(task_or_list, name=?, profile=?, output_schema=?, model=?)` | **`repl`** | Spawn child agent(s), block until done (ultra-mode roots; requires `[delegation] sdk_enabled`). `model=` pins the child's model per request — e.g. a haiku-class id for cheap fan-outs. Blocks the cell — for long-running children run it in a background cell (a user message mid-call backgrounds it; a Stop \u002F cell interrupt cancels the children) |\n| `host.agents.*` \u002F `host.skills.*` | **`repl`** | Profile and skill CRUD — load `customize` skill |\n| `host.submit_output(output, completion_bullets=[...])` | **`repl`** | Submit your structured result when your task carries an OUTPUT SCHEMA section (required before completing). Build the dict in-kernel — the payload rides the host-call wire, not your prose; on a validation\u002Freview bounce, mutate the dict in memory and resubmit (replaces the recorded output) |\n| `host.compute.*` | **`repl`** | Remote job submit\u002Fwait — load the compute skill it names |\n| `host.artifacts(...)` | `python` | Filtered artifact search (wraps the join above) |\n| `host.artifact_path(vid)` \u002F `host.artifact_marker(vid)` | `python` | Resolve a version_id to a readable path \u002F marker |\n| `host.lineage[vid]` | `python` | `{code, messages, env, inputs}` for one version |\n| `host.llm(prompt_or_list, model=?, ...)` | `python` | Single-turn completion via the host's API client. Omitting `model=` uses the Haiku-class kernel default (via `[llm] kernel_default_model`); for harder reasoning pass `model=host.current_model()` — never hardcode a literal model id (they go stale) |\n| `host.credentials.list()` \u002F `.get(name)` | `python` | User-configured credential metadata |\n| `host.mcp(server, method, **kw)` | **`repl`** | MCP\u002Fconnector call — only exists in the `repl` tool; pass results to `python`\u002F`r` via `.\u002Fhandoff\u002F*.json` |\n\nThe `repl` tool and the `python` tool are separate processes that share\nonly the workspace directory — move data between them via\n`.\u002Fhandoff\u002F*.json`, not variables.\n",{"name":5,"license":15,"description":7},"main","e71245365df78e06cd166602744bf2491f79910e","research-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Fself-awareness",0,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS","https:\u002F\u002Fpengqianhan.github.io\u002FAI-Human-Research-OS\u002F","https:\u002F\u002Fgithub.com\u002Fpengqianhan","Ph.D. candidate of The University of Auckland, New Zealand","2026-11-14 18:08:16",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add pengqianhan\u002FAI-Human-Research-OS --skill self-awareness",{"tool":39,"command":40},"bunx","bunx skills add pengqianhan\u002FAI-Human-Research-OS --skill self-awareness",{"tool":42,"command":43},"pnpm","pnpm dlx skills add pengqianhan\u002FAI-Human-Research-OS --skill self-awareness","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fclaude-science-skills\u002Fskills\u002Fself-awareness\u002FSKILL.md",true,[],[48,56,63,70,77,84],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":19},"01M74MWYTHT498SVMN823896T7","workflow-skill-creator","Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say \"make this a skill\", \"create a skill from what we just did\", \"package this workflow\" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).","MIT","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fworkflow_skill_creator\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M74MWWVCFZ8HQS8840F1V1FT","uv","Checks whether the uv Python package manager is installed and installs it if missing. Ensures uv is on PATH. Use when another skill requires uv as a prerequisite.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fuv\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M74MWRH5BTSWZ2C5PCTD5FV9","uniprot-database","Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funiprot_database\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M74MWN1D3ER9F1FZPWH8HQC1","unibind-database","Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED\u002FFASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funibind_database\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M74MWHHB5QZY3NR2FYA1EK4N","ucsc-conservation-and-tfbs","Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fucsc_conservation_and_tfbs\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M74MWERYHSY87P21FDMVE6BA","string-database","Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fstring_database\u002FSKILL.md",[],[]]