boost-skill-cli 1.0.366__tar.gz → 1.0.368__tar.gz

{boost_skill_cli-1.0.366 → boost_skill_cli-1.0.368}/.gitignore

@@ -223,7 +223,34 @@ boost_cli/_version.py
 # BMAD Method — generated artifacts (provisioned via `boost bmad`, not source)
 _bmad/
 _bmad-output/
-.claude/skills/bmad-*/
+
+# Dogfooding boost inside boost's own checkout. Running `boost install --local`
+# here to test a change materializes into exactly the layout the CLI is
+# documented to produce — a project lock, one dotdir per agent target, and a
+# context file per agent with no rules dir — and every one of them lands
+# untracked in this repo. That made a dirty working tree the normal state and
+# put generated content one `git add -A` away from being committed as source.
+#
+# The canonical store lives under $HOME, so nothing here is the store itself;
+# these are only this repo's project-scoped copies.
+.boost/
+.claude/skills/
+.cursor/
+.gemini/
+.windsurf/
+
+# Per-agent context files boost writes for a project-scoped rule
+# (rules.CONTEXT_FILES). CLAUDE.md is *source* and stays tracked; these two are
+# generated — GEMINI.md carries `boost:rule:*` markers, and CLAUDE.local.md is
+# the local-scope half that was never meant to be shared.
+/GEMINI.md
+/CLAUDE.local.md
+
+# Workflows materialize into .claude/commands/, which is also where this repo
+# keeps its own committed slash commands (roadmap-loop.md), so the directory
+# cannot be ignored wholesale. Named entries are the honest fix until the two
+# have somewhere separate to live; add a line when you install another.
+.claude/commands/ship-it.md
 
 # mutmut mutation-testing workspace (regenerated by `mutmut run`)
 mutants/

{boost_skill_cli-1.0.366/boost_skill_cli.egg-info → boost_skill_cli-1.0.368}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.0.366
+Version: 1.0.368
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License: GNU GENERAL PUBLIC LICENSE

{boost_skill_cli-1.0.366 → boost_skill_cli-1.0.368}/boost_cli/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.0.366'
-__version_tuple__ = version_tuple = (1, 0, 366)
+__version__ = version = '1.0.368'
+__version_tuple__ = version_tuple = (1, 0, 368)
 
 __commit_id__ = commit_id = None

{boost_skill_cli-1.0.366 → boost_skill_cli-1.0.368/boost_skill_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.0.366
+Version: 1.0.368
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License: GNU GENERAL PUBLIC LICENSE

{boost_skill_cli-1.0.366 → boost_skill_cli-1.0.368}/boost_skill_cli.egg-info/SOURCES.txt

@@ -114,6 +114,7 @@ docs/demo.gif
 docs/demo.html
 docs/demo.tape
 docs/design-roadmap.html
+docs/eval-latest.json
 docs/eval.html
 docs/index.html
 docs/mcp-hub.html
@@ -416,6 +417,7 @@ scripts/mutation_shards.py
 scripts/mutation_weights.json
 scripts/perf_gate.py
 scripts/post_deploy_smoke.py
+scripts/publish_eval_json.py
 scripts/release_guard.py
 scripts/release_preflight.py
 style/README.md

boost_skill_cli-1.0.368/docs/eval-latest.json

@@ -0,0 +1,43 @@
+{
+  "schema": 1,
+  "generated": "2026-08-02T06:00:00Z",
+  "commit": "",
+  "engine": "bm25",
+  "corpus": {
+    "entries": 57,
+    "queries": 36,
+    "judgments": 107
+  },
+  "metrics": [
+    {
+      "name": "recall@5",
+      "value": 0.5069,
+      "floor": 0.45,
+      "headroom": 0.0569
+    },
+    {
+      "name": "recall@10",
+      "value": 0.6019,
+      "floor": 0.55,
+      "headroom": 0.0519
+    },
+    {
+      "name": "MRR",
+      "value": 0.8384,
+      "floor": 0.78,
+      "headroom": 0.0584
+    },
+    {
+      "name": "nDCG@5",
+      "value": 0.692,
+      "floor": 0.64,
+      "headroom": 0.052
+    },
+    {
+      "name": "nDCG@10",
+      "value": 0.714,
+      "floor": 0.66,
+      "headroom": 0.054
+    }
+  ]
+}

boost_skill_cli-1.0.368/scripts/publish_eval_json.py

@@ -0,0 +1,127 @@
+#!/usr/bin/env python3
+"""Reduce an eval run to the small public JSON the docs site and the portfolio read.
+
+    python3 scripts/publish_eval_json.py                    # from evals/baseline.json
+    python3 scripts/publish_eval_json.py --results r.json   # from a fresh run
+    python3 scripts/publish_eval_json.py --check            # verify committed output is current
+
+Why a separate, tiny file rather than serving ``evals/baseline.json`` directly:
+that file is ~120KB of per-query detail for 36 queries across two arms, and a
+page that fetches it to show five numbers would pull all of it. This is under
+1KB, has a stable shape a page can rely on, and carries the one thing the raw
+results do not — the floors from ``scripts/eval_gate.py``, so a reader can see
+not just the score but the bar it had to clear.
+
+The floors are imported from eval_gate rather than copied. A floor that drifted
+between the gate and the published number would be worse than publishing
+nothing: the site would show a threshold CI is not actually enforcing.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+BASELINE = ROOT / "evals" / "baseline.json"
+OUT = ROOT / "docs" / "eval-latest.json"
+
+sys.path.insert(0, str(ROOT / "scripts"))
+from eval_gate import FLOORS  # noqa: E402  (path set above)
+
+# The order the site renders them in: the two recall metrics, then the ranking
+# metrics. Not sorted alphabetically, which would put MRR between them.
+ORDER = ["recall@5", "recall@10", "MRR", "nDCG@5", "nDCG@10"]
+
+
+def _commit() -> str:
+    """Short SHA of the commit these numbers were produced from, or '' if unknown."""
+    try:
+        out = subprocess.run(["git", "rev-parse", "--short", "HEAD"],
+                             cwd=ROOT, capture_output=True, text=True, timeout=10)
+        return out.stdout.strip() if out.returncode == 0 else ""
+    except (OSError, subprocess.SubprocessError):
+        return ""
+
+
+def build(source: Path, generated: str) -> dict:
+    data = json.loads(source.read_text(encoding="utf-8"))
+    primary = data.get("primary", "bm25")
+    means = data["arms"][primary]["mean"]
+
+    metrics = []
+    for name in ORDER:
+        if name not in means:
+            continue
+        floor = FLOORS.get(name)
+        metrics.append({
+            "name": name,
+            "value": round(means[name], 4),
+            "floor": floor,
+            # Headroom is what a reader actually wants: not "0.69" but "0.05
+            # above the bar". Rounded to the same precision as the value so the
+            # arithmetic on the page is checkable by eye.
+            "headroom": round(means[name] - floor, 4) if floor is not None else None,
+        })
+
+    return {
+        "schema": 1,
+        "generated": generated,
+        "commit": _commit(),
+        "engine": primary,
+        "corpus": {
+            "entries": data.get("corpus", {}).get("entries"),
+            "queries": data.get("golden", {}).get("queries"),
+            "judgments": data.get("golden", {}).get("judgments"),
+        },
+        "metrics": metrics,
+    }
+
+
+def main(argv: list[str] | None = None) -> int:
+    ap = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    ap.add_argument("--results", type=Path, default=BASELINE,
+                    help="results JSON to reduce (default: evals/baseline.json)")
+    ap.add_argument("--generated", default="",
+                    help="ISO-8601 UTC timestamp to stamp; default: keep the "
+                         "committed one, so a no-op run is a no-op diff")
+    ap.add_argument("--check", action="store_true",
+                    help="verify the committed file matches a fresh build; exit 1 on drift")
+    args = ap.parse_args(argv)
+
+    if not args.results.exists():
+        print("publish-eval: %s does not exist" % args.results, file=sys.stderr)
+        return 2
+
+    # Default to the committed timestamp so that re-running without new numbers
+    # produces a byte-identical file. Otherwise every scheduled run would commit
+    # a diff of exactly one line and the history would fill with noise.
+    generated = args.generated
+    if not generated and OUT.exists():
+        try:
+            generated = json.loads(OUT.read_text(encoding="utf-8")).get("generated", "")
+        except ValueError:
+            generated = ""
+
+    fresh = build(args.results, generated)
+    text = json.dumps(fresh, indent=2) + "\n"
+
+    if args.check:
+        current = OUT.read_text(encoding="utf-8") if OUT.exists() else ""
+        if current != text:
+            print("ERROR: docs/eval-latest.json is out of date — regenerate with\n"
+                  "    python3 scripts/publish_eval_json.py\n"
+                  "and commit the result.", file=sys.stderr)
+            return 1
+        print("eval-latest.json is up to date.")
+        return 0
+
+    OUT.write_text(text, encoding="utf-8")
+    print("wrote %s (%d bytes)" % (OUT.relative_to(ROOT), len(text)))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())