Alpha Insights BizAdvisor
Senior business-analyst methodology encoded as a Skill, with harness-enforced workflows that produce decision-grade HTML research reports.
Positive: hooks match only the Write tool, writes are scoped to the user workspace, writing to the install directory is prohibited, html_write_guard warns transparently on JSON parse failure, evidence grading and source tracing are explicit, MIT license and attribution clear. Deducted: hooks execute in-repo Python scripts without per-run user confirmation; full hook script behavior is not statically auditable from the shown files; unverified publisher leaves attribution incomplete.
Positive: the state machine, gate table and degradation paths (manual validation when Bash unavailable) are self-consistent; committed unit tests cover stage3-6/7 validators including evidence ledgers, cascade timestamps, and chart value/unit/currency consistency. Deducted: static review cannot execute tests; no CI workflow evidence shown proving third-party passing; hook implementations not shown so failure-feedback quality is only partially inferable.
Positive: trigger conditions, ten research scenarios, tiering, AskUserQuestion checkpoints and language-following rules are explicit; provider-specific XHS scripts removed from the public package with public-search fallback, so core function does not hard-depend on one unreachable service. Deducted: report generation depends on external web search and MCP tool availability; mainland-China reachability is not addressed; boundaries on non-Claude-Code/Codex runtimes are under-specified.
Positive: V4.1.4 versioning with a detailed CHANGELOG (including security-boundary patches and installer fixes), INSTALL_FOR_AGENTS.md contract, dual-platform README install paths, MIT license, clear layered directory structure and progressive disclosure. Deducted: publisher identity unverified; maintenance responsibility and update path rest on an individual; residual references to private adapters/knowledge base remain in the public package.
Positive: clear value proposition, public demo report, well-defined seven-section report structure and evidence-graded output formats that are directly usable. Deducted: static review cannot verify the demo report was genuinely produced end-to-end by this skill; the 60% time-saving claim lacks third-party evidence; outputs still require human review of data correctness.
Positive: committed unit tests for validator key paths, a public demo report, a per-version changelog with factual corrections — multiple evidence types. Deducted: capped at 5 by static calibration; no CI run records or independent reproduction; fact/inference separation between the demo and real runs is incomplete.
- Hooks auto-execute in-repo Python scripts after install; review scripts/harness/hooks/ contents before use.
- Report data relies on external web search; all figures and conclusions must be human-verified and are not suitable for direct investment or due-diligence decisions.
- Publisher identity is unverified; benefit claims such as '60% time savings' lack independent evidence.
- Reachability of some external data sources from mainland-China networks is not declared; core research tracks may be network-constrained.
- Private Xiaohongshu adapters were removed from the public package; Track E capabilities degrade to public-search fallback in the public build.
What does this skill do, and when should you use it?
Alpha Insights is a business research skill for Claude Code-compatible runtimes and Codex Desktop, covering ten scenarios from industry research and competitive analysis to market entry, investment decisions and due diligence. It is not a prompt pack: V4 introduces "Harness Engineering" — a state machine, stage-gate validators and hook scripts that enforce a seven-stage workflow so the AI cannot silently skip steps in long conversations. Research follows MECE, hypothesis-driven and triangulation methodologies, with every conclusion tagged by source and A/B/C/D confidence, ending in a structured HTML report. Authored by Eric Young, MIT-licensed, and includes an original 3A-8 Steps Strategy framework.
On load it runs resume_check.py to detect an in-progress research project. Triggered, it advances through seven stages: Briefing (tier selection), Framing (MECE decomposition + framework matching), Planning (hypotheses + data-source planning, optional interview prep), Research (parallel multi-track search A–G, triangulation, evidence claim ledger), Insights (red/blue-team adversarial review, anti-pattern checks), Report (HTML generation via report_helper.py with ECharts), and Iteration. Every Write triggers stage_gate_hook.py automatically; html_write_guard.py blocks premature report writing before required artifacts exist. The deliverable chain (research_definition → research_plan → evidence_base → insights → report.html) supports cascade updates — changing an upstream conclusion forces incremental downstream sync. All outputs go to workspace/{project_slug}/ under the user's working directory.
- Investors and consultants who need decision-grade industry or competitor reports with graded evidence and source tracing before committing to a decision
- Founders exploring a new market who start from a vague idea and need the skill to sharpen scope through interactive questioning, hypotheses and validation plans
- Due-diligence teams checking a target company, using primary-source path planning (registries, filings) and numeric integrity ledgers
- Strategy leads preparing annual or 3-year plans who want structured decomposition via Five Forces, PESTEL, BMC and similar frameworks
- Researchers supplementing public-data blind spots with expert interviews, using the optional Stage 3.5 to generate interview guides and feed notes back into the evidence base
What are this skill's strengths and limitations?
- Workflow enforced by script validators and hooks rather than prompt instructions alone — resistant to step-skipping in long conversations
- Complete evidence chain: conclusions tagged with source and A/B/C/D confidence; key numbers registered in an Evidence Claim Ledger
- 19 analysis frameworks + 9 methodologies, including the original 3A-8 Steps Strategy; framework usage is transparently announced to the user
- Explicit cascade-update rules: modifying upstream conclusions forces incremental downstream deliverable sync
- Resumable workspaces make long research runs auditable and recoverable
- Frontmatter hooks use ${CLAUDE_PLUGIN_ROOT} and !command dynamic execution — other runtimes need adaptation (official Codex wrappers provided)
- Serious use requires Python 3, working search/scraping tools and network access; failure of Track A blocks research entirely
- Independent quality review (IQR) relies on subagent mechanisms whose behavior may vary across runtimes
- No automated test suite in the repo; report quality depends heavily on configured data sources
- Internal skill files are in Chinese; non-Chinese users rely on its language rule for correct output language
How do you install this skill?
Recommended: ask your AI coding agent — "Install Alpha Insights from this repository. Follow INSTALL_FOR_AGENTS.md exactly." Codex Desktop direct: git clone https://github.com/Ericyoung-183/alpha-insights.git && cd alpha-insights && python3 scripts/install_codex.py --verify. Claude Code-compatible: clone the repo, copy the package to ~/.claude/skills/alpha-insights, then run python3 ~/.claude/skills/alpha-insights/scripts/verify_cloudcode.py --skill-root ~/.claude/skills/alpha-insights. Keep the root SKILL.md frontmatter hooks intact.
How do you use this skill?
After installation, simply ask a business question, e.g. "Analyze the competitive landscape of the EV charging industry in China". The skill auto-identifies the scenario (competitive analysis), matches frameworks (Five Forces + Competitive Positioning), runs multi-track search and generates an HTML report. On first trigger it confirms the report tier (Tier 1 quick scan / Tier 2 topical brief / Tier 3 deep report) and asks 2-4 clarification questions. You can also start explicitly with /skill alpha-insights. Public channels work out of the box; Xiaohongshu, knowledge bases and internal databases are optional extensions skipped automatically when unconfigured.
How does this skill compare with similar options?
The README itself contrasts Alpha Insights with "typical AI analysis": generic output, no source tracing, single data source, silent step-skipping. Alpha Insights answers those with framework-driven decomposition, evidence grading, multi-track triangulation and script-enforced gates — positioning itself as a replacement for senior-analyst desk research rather than another prompt pack.