swe-workflow-skills 0.11.2 → 0.11.4

package/VERSION

@@ -1 +1 @@
-0.11.2
+0.11.4

package/catalog.json

@@ -253,7 +253,7 @@
     },
     {
       "name": "research-methodology",
-      "description": "Run DS/ML research with defensible claims — framework and rigor, data understood before claims, method decision records, claims traced to evidence, sources verified. Production serving → ml-model-deployment. Triggers: research project, thesis, plan the research track, is this finding solid, can I claim this, why this method, literature review, verify these sources, deep research citations.",
+      "description": "Run DS/ML research with defensible claims — framework and rigor, data understood before claims, method decision records, claims traced to evidence, sources verified, and the write-ups that report them. An experiment or results write-up, even a README, is here, not project-documentation. Production serving → ml-model-deployment. Triggers: research project, thesis, plan the research track, is this finding solid, can I claim this, why this method, literature review, verify these sources, deep research citations, write up the results, experiment README, I can't follow these findings.",
       "path": "research-methodology/SKILL.md"
     },
     {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "swe-workflow-skills",
-  "version": "0.11.2",
+  "version": "0.11.4",
   "description": "A curated library of Claude Code Agent Skills covering the full software lifecycle, with orchestrator-routed activation that scales past the skill-listing budget.",
   "type": "module",
   "bin": {

package/skills/research-methodology/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: research-methodology
-description: "Run DS/ML research with defensible claims — framework and rigor, data understood before claims, method decision records, claims traced to evidence, sources verified. Production serving → ml-model-deployment."
-when_to_use: "Triggers: research project, thesis, plan the research track, is this finding solid, can I claim this, why this method, literature review, verify these sources, deep research citations."
+description: "Run DS/ML research with defensible claims — framework and rigor, data understood before claims, method decision records, claims traced to evidence, sources verified, and the write-ups that report them. An experiment or results write-up, even a README, is here, not project-documentation. Production serving → ml-model-deployment."
+when_to_use: "Triggers: research project, thesis, plan the research track, is this finding solid, can I claim this, why this method, literature review, verify these sources, deep research citations, write up the results, experiment README, I can't follow these findings."
 ---
 
 # Research Methodology
@@ -107,6 +107,28 @@ Write findings at the strength the evidence has:
   only after Step 2's distributions, Step 3's instrument check and the
   confounders are done. Until then it is a provisional observation, worded as one.
 
+**Write each artifact for a reader who wasn't in the session.** An experiment
+README, a results section or a method record is read later by the project owner,
+with nothing else open, and must make the work clear without the notebook or the
+code:
+- **Open with what was found, in plain words**, and what it changes. Provenance
+  (commit, data version) follows the summary; it doesn't lead.
+- **Say what was run**, in a few lines: what it modelled or predicted, on which
+  data, the method and its settings (algorithm, hyperparameters, thresholds), and
+  why that method, with a link to its method decision record. Metrics without the
+  run behind them can't be read. If the material doesn't say, ask; don't leave it out.
+- **Keep each question, its result and what it means for the project together**,
+  in the same row or paragraph, not a table followed by a separate list of
+  implications.
+- **Link shared vocabulary; spell out the rest.** Glossary terms, the methodology
+  and other project-wide definitions have one home: link to it rather than
+  restating it. Shorthand with no home (a notebook symbol, a step code, a decision
+  named by date) is written out, or asked about.
+- **Revising a dense document adds what the work is missing.** Replacing a
+  repeated definition with a link is fine; cutting what was done or what it means
+  is not. Pre-registered text stays verbatim.
+- Don't add qualifiers the output doesn't carry, such as a CI level or a test name.
+
 Under deadline, write the supported part now and mark the rest provisional. The
 minimal check a striking subgroup claim needs (group counts, per-group metric
 with intervals, one confounder cross-tab) is usually an hour, not a revision.

package/skills/research-methodology/evals/evals.json

@@ -53,6 +53,37 @@
         "Names the minimal data check the roof claim needs (e.g. roof-class counts and recall per stratum with intervals) and says it fits before the deadline",
         "Rejects 'add the data description in the next revision' because a claim drafted without the data check tends to survive into the final text"
       ]
+    },
+    {
+      "id": 2,
+      "prompt": "Write the Results section of experiments/AE04_matching/README.md from my notebook output below. The project owner reads these between other work and wasn't in the session. Terms are already in docs/glossary.md, so keep it tight.\n\nPre-registered: H1 R@50m >= 0.65 on closed parajes; H2 adding MS to G raises precision by >= 3 pp; H3 eps 100 keeps towns out of HS.\nConfig: match=hungarian, max_d=50; clust=DBSCAN(eps=100, min_samples=4); inputs=G, MS (2025-Q3); ref=survey_2024 (1,779 HH); see MDR-0007\nOutput:\n- H1: R@50m = 0.71 [0.66, 0.76], n_cp = 41 -> supp\n- H2: dPrec(MS u G - G) = +2.1 pp [-0.4, 4.6] -> inconcl\n- H3: eps 100 merges M-masked towns into 39 HS -> not supp; A.7 pending\n- l = 225 m (grid), imposib. excluded per 2026-10-22",
+      "pressure": [
+        "brevity",
+        "glossary_exists"
+      ],
+      "expected_behavior": "Writes the section for a reader who was not in the session: a plain-language summary of what was found first; a few lines on what was run (the matching and clustering methods with their settings, the inputs and the reference data, the method's reason linked to its MDR); then each hypothesis with its question, result and meaning for the project together. Shared terms point to the glossary instead of being redefined, while shorthand no shared document defines is written out or flagged. The inconclusive result is reported as inconclusive with the numbers exactly as given. Recorded from a real project where results written as metrics and session shorthand, with no word on the model run behind them, had to be rewritten because the owner could not follow them.",
+      "assertions": [
+        "Opens with a short plain-language summary of what was found, before the per-hypothesis detail",
+        "Says in a few plain lines what was run (the matching and clustering methods with their settings, the inputs and the reference data) and links the method's reason to MDR-0007, rather than reporting only the metrics",
+        "For each hypothesis, states the question asked, the result, and what it means for the project, next to each other",
+        "Points shared terms to the glossary rather than redefining them, and writes out or flags the shorthand no shared document defines (A.7, the 2026-10-22 decision)",
+        "Reports H2 as inconclusive, not as an effect, and keeps every number as given without inventing values"
+      ]
+    },
+    {
+      "id": 3,
+      "prompt": "I can't follow experiments/AE02_t0_products_eda/README.md. It's about 300 lines: the pre-registration (rules written before the run) comes first, then results as lists of numbers like 'cov@50 = 0.94 (med 6 m); rand 0.05; dCov(MS|G) +5.3 pp [3.3, 7.5] < 5 -> inconcl; FP/HH G 5.0 vs MS 2.65; M share 47-53%'. Each question is about 150 lines from its answer. We also have docs/glossary.md. Fix the README so I can follow it; tell me what you'd change.",
+      "pressure": [
+        "dry_instinct"
+      ],
+      "expected_behavior": "Treats the complaint as too little explanation of the work, not too much text: adds a plain summary, what the experiment ran (data, method and settings, what it set out to measure) and the meaning of each result, puts each question next to its answer, and leaves the pre-registration verbatim. Shared terms stay in the glossary and are linked, not copied in. Recorded from a real project where the owner could not follow results given as numbers and acronyms with no account of the run behind them.",
+      "assertions": [
+        "Adds explanation (a plain summary, what each result means) rather than shortening or cutting content",
+        "Adds, or asks for, what the experiment ran (the data, the method and its settings, what it set out to measure) so the README can be followed without the notebook",
+        "Puts each question next to its answer and its meaning",
+        "Leaves the pre-registration unchanged, or states that it must stay verbatim, while restructuring around it",
+        "Treats docs/glossary.md as the home for shared terms, linking to it rather than adding a second definition in the README"
+      ]
     }
   ]
 }