菜单
Searches the web for documentation on an unknown merger tree format. Use when kdb-lookup finds no matching format in the KDB and the input format must be identified from external sources.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Guides drafting or adapting a format-specific conversion driver for the SAGE merger tree converter. Use when no existing driver matches the input format and a new driver must be written in assets/ before the test conversion can run.
Provides field-level guidance for mapping input merger tree fields to the SAGE LHaloTree schema (HDF5 or binary). Use during web discovery or KDB entry creation when individual field mappings, unit conversions, or pointer reconstruction decisions need to be made.
Independently verifies the semantic validation plots and plotting code against the mandatory checklist. Use after semantic-validation generates the seven plots, before presenting results to the user. Operates independently - does not rely on the main agent's assessment.
Generates the seven mandatory semantic validation plots of the converted output merger trees. Use after a full conversion run in Stage 3 to assess whether the converted trees are physically plausible.
Runs the six structural checks on a converted SAGE LHaloTree HDF5 file. Use after any conversion run — test or full — to verify the output file is structurally correct before proceeding.
Adds a new format driver and schema mapping JSON to the KDB after a successful conversion of a previously unknown format. Use in Stage 4 when the input format was not found in the KDB at session start.
| name | web-discovery |
| description | Searches the web for documentation on an unknown merger tree format. Use when kdb-lookup finds no matching format in the KDB and the input format must be identified from external sources. |
This skill is invoked when kdb-lookup finds no full match in format-database/.
The goal is to locate sufficient external documentation to produce a complete schema
mapping that can be confirmed by the user at Gate G1.
Two path prefixes are used in this skill:
.ai/skills/web-discovery/references/<file> — files in this skill's own references/ subfolder.reference/<file> (project root) — files in the project-root reference/ directory (e.g. reference/sage_lhalotree_hdf5_schema.md).Construct search queries combining:
Query forms to try (in priority order):
"<halo_finder>" "<tree_tool>" output format columns"<halo_finder>" merger tree file format documentationsite:github.com "<halo_finder>" "<tree_tool>"Source priority (highest to lowest) — see .ai/skills/web-discovery/references/source_credibility.md:
Stop searching once you have found a complete field list and a pointer semantics description from a Tier 1 or Tier 2 source.
Once field information is found, read reference/sage_lhalotree_hdf5_schema.md (project root) and
for each mandatory SAGE field, identify the closest matching source field. Note:
Spin as spin parameter vs. specific
angular momentum vector — use the schema-mapping skill to resolve these)Invoke the schema-mapping skill for all field-level and pointer-level decisions.
Write the resulting mapping to assets/proposed_mapping_<format_id>.json, following
reference/format_database_template.json (project root) exactly. The format_id is
<halo_finder_lower>_<tree_tool_lower>_<file_format_lower> with underscores.
The draft must include:
field_map for every mandatory SAGE fieldpointer_logic section describing how pointers are reconstructedreferencesPresent the draft mapping to the user with a summary of:
Incorporate user corrections, update assets/proposed_mapping_<format_id>.json, and
present again. Repeat until the user issues an explicit confirmation (Gate G1).