菜单
Produce a low-fidelity market-brief.md stub from codebase signals (README, manifests, ADRs, git tags, deploy config) during /codify. Owned by product-keeper. Confidence defaults to low; real market data is never inferable from code.
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-market-brief-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-market-brief-from-code |
| description | Produce a low-fidelity market-brief.md stub from codebase signals (README, manifests, ADRs, git tags, deploy config) during /codify. Owned by product-keeper. Confidence defaults to low; real market data is never inferable from code. |
| version | 0.1.0 |
| user-invocable | false |
| model | sonnet |
| allowed-tools | Read, Write, Grep, Glob, Bash |
| 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| deprecated_note | #434 ProductOS realignment — superseded by the command model; retained for Phase E reference, not installed |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. Produces specification/market-brief.md at {stm_base}/{issue}/evidence/codify/proposals/specification/market-brief.md.
Emit a low-fidelity stub of the greenfield market brief using only codebase-visible signals. In /specify the market-analyst agent produces this artifact from web research; in /codify the codebase is the sole window. Expect thin evidence — README framing, manifest descriptions, ADR business context, git tag cadence, deploy targets — none of which surface TAM/SAM/SOM, competitive landscape, pricing strategy, or market sizing.
This skill is explicitly a stub producer. Confidence should almost always resolve to low. The artifact's most important section is knowledge_gaps — it lists what could not be inferred so downstream /codify flows file R4 entries in resolution-trace and humans can enrich the brief later.
| Field | Required | Description |
|---|---|---|
scan_index_path | yes | Path to scan-index.json produced by scan.py |
stm_base | yes | STM root (resolved from .garura/core/config.yaml stm.base-path) |
issue | yes | Issue number driving /codify |
output_path | yes | Target path — typically {stm_base}/{issue}/evidence/codify/proposals/specification/market-brief.md |
decision_manifest_path | yes | Where per-decision provenance is recorded |
ltm_context | optional | Resolved LTM pointers from the Resolution Protocol (prior market briefs as templates) |
resolution_trace_path | yes | Where R1–R4 resolution decisions are recorded |
Validate inputs. Confirm scan_index_path exists and is readable; confirm stm_base/{issue}/ is writable; confirm output_path parent directory exists (create if absent).
Walk R1 → R4 for any inferred market claim. R3 may consult for prior market briefs as structural templates (never as evidence). Write resolution decisions to . Claims with no KB or web backing land as R4 entries and feed .
core/components/memory/resolution_trace_pathknowledge_gapsSignal extraction from scan-index.json.
docs.readme_preview — pull positioning language, target-user phrasing, feature bullets. Scan for phrases like "for {audience}", "helps you {verb}", "alternative to X", "similar to Y". Capture literal quotes with line references — never paraphrase positioning and treat the paraphrase as evidence.docs.adrs — scan each ADR preview for business-context framing (market rationale, positioning decisions, competitive reasoning). Technical ADRs are irrelevant here and must be skipped.manifests[*].name / manifests[*].description — elevator-pitch level framing; one line each. A manifest description is rarely richer than a tagline.git.tags_recent — release cadence as a crude maturity proxy (dense tags → active; sparse or none → early/dormant). Never a market signal — record as context only.config_files.deploy — deployment targets hint at distribution model (self-hosted binaries vs SaaS vs library vs CLI). Never a pricing or go-to-market signal; record as distribution hypothesis only.frontend_detection and entry_points — presence of a web UI vs CLI-only vs API-only hints at buyer-user shape. Record as hypothesis only.Author market-brief.md with YAML frontmatter (the meta block) + markdown body. Body sections, in order:
# Product positioning — one paragraph drawn from README framing. If README is absent or silent, state so explicitly and route to knowledge_gaps.# Target users — bullet list inferred from README language (personas named, use cases described). Each bullet tagged with [evidence: readme_preview L{n}] or [inferred].# Value proposition — what the README or manifest description claims the product does for whom. One to three bullets.# Competitive context — thin by default. Only populate if README literally names competitors or positions against a category; otherwise a single line: "No competitive context inferable from codebase signals. See knowledge_gaps."# Knowledge gaps — always substantial. Itemize at minimum:
source_type: uninferable_from_code to drive R4 entries in resolution-trace.Record decisions in decision manifest. Every inferred statement in the body carries an entry with: claim, signal(s) cited, confidence (low | medium — never high in this skill), source_type (inferred_from_code | uninferable_from_code). Low-confidence statements dominate.
Primary artifact: market-brief.md at output_path with YAML frontmatter:
---
meta:
source_type: "inferred_from_code"
evidence:
- "{scan_index_path}#docs.readme_preview"
- "{scan_index_path}#manifests[*].description"
- "{scan_index_path}#git.tags_recent"
confidence: "low" # expected default — only raise to "medium" if README is genuinely rich and names real market context
learning_category: "product"
sub_category: null
tier: 2
note: "Low-fidelity stub. Market context is not inferable from code alone. See knowledge_gaps section for items requiring human input or web research."
---
followed by the markdown body described in Process step 4.
Decision manifest at decision_manifest_path — one entry per inferred body statement.
Resolution trace at resolution_trace_path — R1–R4 entries for every claim, with R4 entries dominating.
Never inflate confidence to paper over thin evidence. A medium rating requires README content that explicitly states positioning, target users, and at least one named competitor or market category. Absent all three, stay at low. A missing README forces low with a single-line body pointing everything to knowledge_gaps — do not fabricate positioning from code structure alone.
scan_index_missing — scan_index_path does not exist or is malformed.ltm_resolution_failed — Resolution Protocol could not complete (e.g., memory path unreadable); proceed with R4-only trace and flag in meta.note.insufficient_signal — no README, no manifest descriptions, no business ADRs. Proceed with confidence: low, a minimal body, and an exhaustive knowledge_gaps section. Do not fail the skill — the stub is the point.output_path, decision_manifest_path, resolution_trace_path.aggregate-codify-proposals' job.