菜单
For every domain in the upstream domain-selection proposal, derive a product-scoped research Markdown file from scan-index.json evidence plus the canonical domain KB, carrying the five KB-extension sections with inline knowledge_gap markers wherever code evidence is absent.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/kapilvirenahuja/garura --skill infer-research-from-code命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | infer-research-from-code |
| description | For every domain in the upstream domain-selection proposal, derive a product-scoped research Markdown file from scan-index.json evidence plus the canonical domain KB, carrying the five KB-extension sections with inline knowledge_gap markers wherever code evidence is absent. |
| version | 0.1.0 |
| user-invocable | false |
| model | sonnet |
| allowed-tools | Read, Write, Grep, Glob |
| deprecated | true |
| deprecated_note | #434 ProductOS realignment — superseded by the command model; retained for Phase E reference, not installed |
Called by product-keeper during /codify, strictly after
infer-domain-selection-from-code has produced specification/domain-selection.yaml.
Produces one research/{domain}.md per selected domain at
{stm_base}/{issue}/evidence/codify/proposals/research/{domain}.md.
In greenfield, research/{domain}.md is copied from the canonical KB during
/specify Stage 3 and extended by research agents. In /codify the codebase
is the evidence surface — no prior product LTM, no web research. This skill
derives each domain's research document from scan-index.json signals attributed
to that domain, the canonical KB entry (when one exists), and any ADR / README
material matching the domain's search patterns. Every output carries the five
KB-extension sections from core/components/memory/standards/rules/kb-extension.md;
where scan evidence cannot populate a section, it is emitted with an inline
knowledge_gap: true marker so /garura:enrich surfaces the gap rather than silently
resolving it. Tier-2 inference, learning_category: domain, sub_category: null.
Receive from the /codify play orchestrator via JSON contract. All paths absolute.
| Field | Required | Description |
|---|---|---|
scan_index_path | yes | scan-index.json produced by scan.py. |
stm_base | yes | STM root from stm.base-path in .garura/core/config.yaml. |
issue | yes | Issue number (STM namespace). |
related_proposal_paths | yes | MUST contain the absolute path to specification/domain-selection.yaml. Missing → missing_related_proposal. |
kb_domain_dir | yes | core/components/memory/knowledge/domain/. Missing → kb_domain_dir_missing. |
output_dir| yes |
{stm_base}/{issue}/evidence/codify/proposals/research/. One {domain}.md per selected domain lands here. |
decision_manifest_path | yes | {…}/proposals/research/decision-manifest-infer-research.yaml. |
resolution_trace_path | yes | {…}/proposals/research/resolution-trace-infer-research.yaml. |
ltm_context | yes | Resolution Protocol context (product_base, kb_base, resolved LTM paths). |
scan_index_path exists and parses as JSON with manifests, patterns,
config_files, docs, git. Missing → scan_index_missing.related_proposal_paths includes an existing domain-selection.yaml.
Missing → missing_related_proposal.kb_domain_dir exists with ≥1 non-underscore .md file. Missing →
kb_domain_dir_missing.output_dir is creatable. Failure → output_path_unwritable.Per core/components/memory/standards/rules/resolution.md. Write every probe
to resolution_trace_path with source, path, outcome, payload.
"codify-bootstrap".{product_base}/research/{domain}.md per
selected domain. If present AND LOCKED, propose only when scan evidence
materially conflicts; non-conflicts emit alignment_confirmed: true in
frontmatter and skip the body.domain-selection.yaml:
inferred_domain_new: false → read {kb_domain_dir}/{domain}.md to inherit
canonical feature IDs, search patterns, and five-section structure.
inferred_domain_new: true → no KB base; structure directly from
kb-extension.md and emit knowledge_gap: true on every section without
a code signal./garura:enrich to a human.Replay the signal → domain mapping used by infer-domain-selection-from-code,
persisting per-signal evidence under the owning domain:
patterns.framework_idioms tagged for this domain (express → backend-api).patterns.naming_suffix_counts whose suffix is typical for the domain
(*Controller/*Service → backend-api; *Screen/*Page → frontend-web).patterns.test_framework_signals whose test files sit under paths
implicated by the domain's source files.config_files.* associated with the domain (Dockerfile → containerized
runtimes; .github/workflows → engineering-experience; helm → deployment).docs.adrs whose title contains the domain's search-pattern tokens (from
KB base for canonical domains; derived from slug + triggering readme
excerpt for new domains).docs.readme_preview sections whose heading or surrounding paragraph
matches the domain's search patterns — record matched span and line.manifests deps mapped to the domain (Stripe → payments; Auth0/Clerk
→ user-management; etc.).git.churn_top is used ONLY for Experiential; never as membership evidence.Every item carries a concrete evidence path (JSON pointer or file path) and, where applicable, a literal excerpt ≤ 200 chars.
research/{domain}.md — five required sectionsPer selected domain emit Markdown: YAML frontmatter (Section 5), short prose
preamble, five KB-extension sections in order. Each section states
inferred: true and emits inline knowledge_gap: true bullets rather than
fabricating content when evidence is absent.
*Controller suffix present → backend-api included"). knowledge_gap: true
is NOT permitted here; a domain without inclusion justification would not
be in selected_domains.- knowledge_gap: true — no measurable success criteria observable from scan; defer to /garura:enrich.knowledge_gap: true with pointers to insufficient-but-present signals
(ADR titles, churn clusters) as leads.git.churn_top
and cross-domain imports in manifests / entry_points. If no cross-tree
signal exists, emit - (none observed) — a valid finding, NOT a
knowledge_gap.git.churn_top
entries with earliest commit > 180 days old). Usage count bootstraps to
0; Last promoted: never. If no ADR / README / churn match, emit
knowledge_gap: true on Scenarios-observed and Common-mistakes sub-bullets.Every file begins with:
---
source_type: "inferred_from_code"
evidence:
- "<scan-index JSON pointer or file path>"
- "..."
confidence: "high" | "medium" | "low"
learning_category: "domain"
sub_category: null
tier: 2
domain: "<slug>"
inferred_domain_new: <bool>
alignment_confirmed: <bool> # true only when R2 hit LOCKED with no conflict
knowledge_gap_sections: # explicit catalog of gap-marked sections
- "Success Criteria"
- "Failure Scenarios"
---
Confidence uses the same tiers as infer-domain-selection-from-code, computed
per-domain from the signal count in THIS research file.
Write every research/{domain}.md first, then the decision manifest, then the
resolution trace. The manifest lists every section-level inference — one entry
per (domain, section) derived from evidence AND one entry per (domain, section)
emitted as knowledge_gap: true. Per Decision Surfacing Discipline each entry
carries decision_id, decision_type, tier (low for knowledge_gap, mid/high
for evidenced), grounding_source, recommendation, alternatives,
chosen_reason. Nothing silently resolved.
Primary artifacts — research/{domain}.md (one per selected domain):
YAML frontmatter meta block + prose preamble + five KB-extension sections
(Inclusion, Success Criteria, Failure Scenarios, Cross-Tree Refs, Experiential).
Sections with no code evidence carry inline knowledge_gap: true markers;
sections with evidence cite concrete scan-index pointers or file paths.
Decision manifest — decision-manifest-infer-research.yaml: standard
Garura shape (schema_version, skill, generated_at, decisions[]); one
entry per (domain, section) inference or knowledge_gap.
Resolution trace — resolution-trace-infer-research.yaml: R1..R3 probes
with source, path, outcome, payload. R4 omitted.
status: success
artifact_paths: [ "<output_dir>/<domain>.md", "..." ]
decision_manifest_path: "<decision_manifest_path>"
resolution_trace_path: "<resolution_trace_path>"
research_file_count: <int>
knowledge_gap_section_count: <int>
overall_confidence: "high" | "medium" | "low"
alignment_confirmed_count: <int>
Emit status: failure with what_failed, detail, and evidence (offending
path / field). Codes:
missing_related_proposal — related_proposal_paths lacks an existing
domain-selection.yaml.kb_domain_dir_missing — kb_domain_dir absent or empty.scan_index_missing — scan_index_path absent or not valid JSON.insufficient_signal — domain-selection declares domains but every
per-domain evidence bundle is empty; would produce pure knowledge_gap
stubs. Fall back to /garura:enrich.output_path_unwritable — output_dir cannot be created.scan_index_path, kb_domain_dir, and related_proposal_paths..garura/product/research/ is handled downstream by /garura:enrich.scan.py, consult the web, or modify the canonical KB even
when inferred_domain_new: true — new canonical entries are created only
after human promotion.