Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› rt-execution-bridge
rt-execution-bridge
GitHub将实证分析设计转化为可审计的估计结果。支持DiD、IV、RDD等方法的执行与诊断，通过StatsPAI/Stata工具生成报告，确保引用合规并处理期刊格式要求。
Trigger Scenarios
需要运行实际估计值及诊断指标时
需回应审稿人关于方法偏差的质疑时
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill rt-execution-bridge -g -y
SKILL.md
Frontmatter
{
"name": "rt-execution-bridge",
"description": "Use when an empirical analysis should be RUN and audited, not just advised — DiD, IV, RDD, synthetic control, DML, multiple-testing, sensitivity. Maps the design to the concrete StatsPAI \/ Stata MCP tools in this environment and reports the fitted, audited number. Defers result placement and house style to the target journal's pack."
}
Execution Bridge (rt-execution-bridge)
Close the last mile: turn "you should use a heterogeneity-robust DiD / weak-IV-robust CI /
multiple-testing correction" into an actual fitted, audited estimate. Full map +
orchestration spine + validated worked-examples (DiD / IV / RDD / synthetic-control / DML):
shared-resources/empirical-methods/execution-with-mcp.md.
When to trigger
- You have data and a design and need the actual estimate + its diagnostics.
- A reviewer objection (real or simulated) needs an executed answer (e.g. "TWFE is biased under staggered adoption" → run Callaway–Sant'Anna + Goodman-Bacon).
The spine (always)
detect_design→preflight/recommend→ fit withas_handle=true.audit_result(result_id)— enumerate the checks the design still owes; run eachsuggest_functionit names.- Design-specific sensitivity from the handle (
honest_did_from_result,sensitivity_from_result,evalue_from_result). bibtex(keys=[…])for citations — never invent references.
Design → tools (summary; full table in the canonical doc)
- Staggered DiD:
callaway_santanna/sun_abraham+bacon_decomposition+honest_did_from_result. - IV:
iv+effective_f_test+anderson_rubin_ci(weak-IV-robust). - RDD:
rdrobust+rddensity/mccrary_test. - Synthetic control:
synth/sdid+ placebo inference. - DML / high-dim:
dml+dml_diagnostics(overlap) +oster/sensemakr. - Multiple testing / inference:
romano_wolf,wild_cluster_bootstrap,twoway_cluster. - Exhibits:
etable/did_summary_to_latexstraight from the handle.
Hard rules
- Run, don't claim — every reported estimate/CI/F/bound traces to a tool call.
bibtexis the only citation source.- Method here, placement there — where the result goes (body vs. appendix, page limit,
house table style) is the target pack's skills +
official-source-map.md. - Degrade honestly — if StatsPAI/Stata are not connected, adapt the
code/skeleton and flag any unverified number.
Output format
【Design】DiD / IV / RDD / SCM / DML / …
【Estimate】point [CI] (estimator)
【Key diagnostic】(first-stage/effective F, pre-trends p, overlap, placebo p, …)
【Audit gaps run】…
【Magnitude】interpretable units
【Next】rt-submission-readiness / the pack's tables-figures skill
Version History
- 1839142 Current 2026-07-05 14:19
