boost-skill-cli 1.2.71__tar.gz → 1.2.73__tar.gz

boost_skill_cli-1.2.73/.lighthouserc.json

@@ -0,0 +1,44 @@
+{
+  "ci": {
+    "collect": {
+      "startServerCommand": "python3 scripts/serve_docs.py --port 8099 --bind 127.0.0.1",
+      "startServerReadyPattern": "Serving HTTP",
+      "url": [
+        "http://127.0.0.1:8099/docs/index.html",
+        "http://127.0.0.1:8099/docs/roadmap.html"
+      ],
+      "numberOfRuns": 3,
+      "settings": {
+        "chromeFlags": "--no-sandbox --headless=new"
+      }
+    },
+    "assert": {
+      "assertMatrix": [
+        {
+          "matchingUrlPattern": "index\\.html$",
+          "aggregationMethod": "median-run",
+          "assertions": {
+            "categories:performance": ["error", { "minScore": 0.9 }],
+            "categories:accessibility": ["error", { "minScore": 0.95 }],
+            "categories:best-practices": ["error", { "minScore": 0.9 }],
+            "categories:seo": ["error", { "minScore": 0.95 }]
+          }
+        },
+        {
+          "matchingUrlPattern": "roadmap\\.html$",
+          "aggregationMethod": "median-run",
+          "assertions": {
+            "categories:performance": ["error", { "minScore": 0.8 }],
+            "categories:accessibility": ["error", { "minScore": 0.95 }],
+            "categories:best-practices": ["error", { "minScore": 0.9 }],
+            "categories:seo": ["error", { "minScore": 0.95 }]
+          }
+        }
+      ]
+    },
+    "upload": {
+      "target": "filesystem",
+      "outputDir": ".lighthouseci"
+    }
+  }
+}

{boost_skill_cli-1.2.71/boost_skill_cli.egg-info → boost_skill_cli-1.2.73}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.2.71
+Version: 1.2.73
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License-Expression: Apache-2.0

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73}/boost_cli/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.2.71'
-__version_tuple__ = version_tuple = (1, 2, 71)
+__version__ = version = '1.2.73'
+__version_tuple__ = version_tuple = (1, 2, 73)
 
 __commit_id__ = commit_id = None

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73}/boost_cli/commands/discovery.py

@@ -96,6 +96,12 @@ def cmd_search(argv):
                    help="rerank the top hits with Claude")
     p.add_argument("--limit", type=util.positive_int, default=15,
                    help="max results (default 15)")
+    p.add_argument("--collapse-near-duplicates", action="store_true",
+                   dest="collapse_dupes",
+                   help="collapse near-identical entries (translations, "
+                        "paraphrases) via dense embeddings — opt-in, needs a "
+                        "built dense index; threshold not yet validated "
+                        "against a real corpus, see the roadmap")
     p.add_argument("--json", action="store_true", dest="as_json",
                    help="machine-readable output")
     args = p.parse_args(argv)
@@ -127,7 +133,9 @@ def cmd_search(argv):
         # retrieve_any, not retrieve: this picks the dense backend when one is
         # built and floors to BM25 otherwise, so the CLI and the MCP server
         # answer from the same engine instead of the CLI being BM25-only.
-        hits, engine = rag.retrieve_any(query, k=max(60, args.limit * 4))
+        hits, engine = rag.retrieve_any(
+            query, k=max(60, args.limit * 4),
+            collapse_near_duplicates=args.collapse_dupes)
         scored = [(h["entry"], h["score"]) for h in (hits or [])]
     else:
         scored = catalog.search(query)

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73}/boost_cli/core/dense.py

@@ -1561,6 +1561,51 @@ def _knn(con: sqlite3.Connection, qblob: bytes,
     return _expand(con, ranked, pool)
 
 
+def entry_vectors(entry_keys: list[tuple[str, str]],
+                  ) -> dict[tuple[str, str], bytes]:
+    """Raw float32 embedding for each entry's first chunk, keyed by ``(tap, path)``.
+
+    Feeds `rag.collapse_near_duplicate_hits`, whose clustering needs one
+    representative vector per *entry* rather than per chunk. The lowest-``cix``
+    chunk is that representative: `rag.chunk` always emits the name+description
+    chunk first, which is enough signal to place a translation or a paraphrase
+    near its original without reading the rest of the body.
+
+    Only a quantized store answers this cheaply: `vec_raw` is a plain
+    rowid-keyed table, so looking vectors up by id is an index probe (see
+    `_knn`'s comment on why the same lookup against a `vec0` table plans as a
+    full scan instead). A legacy float32 store returns ``{}``, and near-dup
+    collapsing degrades to a no-op there — the same way every other feature on
+    that path degrades, rather than paying a full-scan cost per entry.
+    """
+    if not entry_keys:
+        return {}
+    con = _connect()
+    if con is None:
+        return {}
+    try:
+        if not quantized(con):
+            return {}
+        col = _vid_col(con)
+        vids: dict[tuple[str, str], int] = {}
+        for tap, path in entry_keys:
+            row = con.execute(
+                "SELECT %s FROM chunks WHERE tap = ? AND path = ? "  # noqa: S608  col is a literal from _vid_col; tap/path are bound
+                "ORDER BY cix LIMIT 1" % col, (tap, path)).fetchone()
+            if row is not None and row[0] is not None:
+                vids[(tap, path)] = row[0]
+        if not vids:
+            return {}
+        ids = sorted(set(vids.values()))
+        rows = con.execute(
+            "SELECT id, embedding FROM vec_raw WHERE id IN (%s)"  # noqa: S608  interpolates only `?` placeholders; ids are bound
+            % ",".join("?" * len(ids)), ids).fetchall()
+        blobs = dict(rows)
+        return {key: blobs[vid] for key, vid in vids.items() if vid in blobs}
+    finally:
+        con.close()
+
+
 def retrieve(query: str, k: int = 60, kind: str | None = None,
              entries: list[dict] | None = None) -> list[Hit] | None:
     """Top-k dense hits for ``query``, or None to signal 'fall back to BM25'."""

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73}/boost_cli/core/rag.py

@@ -17,6 +17,7 @@ no AI -> the BM25 order is returned as-is.
 """
 from __future__ import annotations
 
+import array
 import contextlib
 import hashlib
 import json
@@ -773,6 +774,91 @@ def dedupe_by_content(hits: list[Hit], limit: int) -> list[Hit]:
     return out[:limit]
 
 
+#: Cosine-similarity floor for two entries to collapse as near-duplicates.
+#:
+#: Deliberately conservative and deliberately unproven. `dedupe_by_content`
+#: above shipped only after one count settled its safety: of 14,153 distinct
+#: bodies, the clusters spanning more than one *name* numbered zero, so
+#: collapsing on content hash could not merge two different skills. No
+#: equivalent count exists yet for near-identical clustering — the measurement
+#: the roadmap card asks for is "over a real corpus, at the chosen threshold,
+#: count clusters spanning more than one meaning" and nobody has run it. Until
+#: someone does, `collapse_near_duplicate_hits` stays reachable only opt-in
+#: (see `retrieve_any`'s `collapse_near_duplicates` parameter), so the default
+#: search path is unaffected by a threshold nobody has validated.
+NEAR_DUPLICATE_THRESHOLD = 0.97
+
+
+def _decode_vector(blob: bytes) -> array.array:
+    """A stored float32 embedding blob as a sequence of floats."""
+    vec = array.array("f")
+    vec.frombytes(blob)
+    return vec
+
+
+def _cosine(a: array.array, b: array.array) -> float:
+    """Cosine similarity of two float32 vectors, 0.0 for any comparison that
+    cannot be trusted: a dimension mismatch (different embedding models) or
+    either side being the zero vector. An ambiguous comparison must never read
+    as a match, the same rule a missing content hash follows above.
+    """
+    if not a or not b or len(a) != len(b):
+        return 0.0
+    dot = sum(x * y for x, y in zip(a, b, strict=True))
+    norm_a = math.sqrt(sum(x * x for x in a))
+    norm_b = math.sqrt(sum(x * x for x in b))
+    if 0.0 in (norm_a, norm_b):
+        return 0.0
+    return dot / (norm_a * norm_b)
+
+
+def collapse_near_duplicate_hits(hits: list[Hit],
+                                  vectors: dict[tuple[str, str], bytes],
+                                  limit: int,
+                                  threshold: float = NEAR_DUPLICATE_THRESHOLD,
+                                  ) -> list[Hit]:
+    """Collapse entries whose embeddings are near-identical, then take ``limit``.
+
+    Runs on the output of :func:`dedupe_by_content`, on the same contract —
+    keep the earliest rank slot, promote a better source into it — for the
+    residual that survives byte-identical dedup: the same skill translated or
+    paraphrased across mirrors. Observed on a real 466-tap install: the query
+    ``exa search`` returned ten rows that were all one skill in Japanese,
+    Chinese, and five English variants, none sharing a body digest, so
+    ``dedupe_by_content`` correctly kept every one of them apart.
+
+    ``vectors`` maps ``entry_key(hit["entry"])`` to a raw float32 embedding
+    blob, looked up by the caller (:func:`boost_cli.core.dense.entry_vectors`)
+    rather than fetched here — this stays pure so it is cheap to test with
+    synthetic vectors, and clustering ``hits`` is O(n * clusters), never
+    O(n^2) against the whole store.
+
+    A hit with no vector — no dense store built, or the entry never got one —
+    is never merged into anything: two unknowns must not collapse into each
+    other, the same rule :func:`dedupe_by_content` follows for a missing
+    content hash.
+    """
+    out: list[Hit] = []
+    reps: list[array.array | None] = []   # parallel to `out`
+    for hit in hits:
+        blob = vectors.get(entry_key(hit["entry"]))
+        vec = _decode_vector(blob) if blob else None
+        merge_at = None
+        if vec is not None:
+            for i, rep in enumerate(reps):
+                if rep is not None and _cosine(vec, rep) >= threshold:
+                    merge_at = i
+                    break
+        if merge_at is None:
+            out.append(hit)
+            reps.append(vec)
+            continue
+        kept = out[merge_at]
+        if source_rank(hit["entry"]) < source_rank(kept["entry"]):
+            out[merge_at] = cast(Hit, {**kept, "entry": hit["entry"]})
+    return out[:limit]
+
+
 def _windowed(hits: list[Hit], terms: Sequence[str]) -> list[Hit]:
     """Window each hit's raw stored snip onto the query terms.
 
@@ -1087,7 +1173,8 @@ def _with_content(hits: list[Hit]) -> list[Hit]:
 
 
 def retrieve_any(query: str, k: int = 60, kind: str | None = None,
-                 entries: list[dict] | None = None
+                 entries: list[dict] | None = None,
+                 collapse_near_duplicates: bool = False,
                  ) -> tuple[list[Hit] | None, str]:
     """Fuse BM25 and dense when both are available; floor to whichever is.
 
@@ -1107,6 +1194,12 @@ def retrieve_any(query: str, k: int = 60, kind: str | None = None,
     Both engines are over-fetched to ``RRF_K`` before fusing. Fusing only the
     top-k of each would discard exactly the candidates the other engine was
     going to promote, which is most of the value.
+
+    ``collapse_near_duplicates`` runs :func:`collapse_near_duplicate_hits` over
+    the result, on top of the byte-identical dedup every branch below already
+    does — off by default (see :data:`NEAR_DUPLICATE_THRESHOLD` for why) and a
+    no-op whenever no dense store is built, since that is the only source of
+    the embeddings the collapse needs.
     """
     from . import dense
     pool = max(k, RRF_K)
@@ -1127,13 +1220,20 @@ def retrieve_any(query: str, k: int = 60, kind: str | None = None,
     # no reason to treat them as one.
     if dense_hits and bm25_hits:
         fused = rrf_fuse([bm25_hits, dense_hits], limit=max(k, RRF_K))
-        return (dedupe_by_content(_with_content(fused), k),
-                "hybrid RRF (BM25 + dense)")
-    if dense_hits:
-        return dedupe_by_content(_with_content(dense_hits), k), "dense vectors"
-    if bm25_hits is not None:
-        return bm25_hits[:k], "BM25 full-content"     # already deduped
-    return None, ""
+        hits, engine = dedupe_by_content(_with_content(fused), k), \
+            "hybrid RRF (BM25 + dense)"
+    elif dense_hits:
+        hits, engine = dedupe_by_content(_with_content(dense_hits), k), \
+            "dense vectors"
+    elif bm25_hits is not None:
+        hits, engine = bm25_hits[:k], "BM25 full-content"     # already deduped
+    else:
+        return None, ""
+    if collapse_near_duplicates and hits and dense.ready():
+        vectors = dense.entry_vectors([entry_key(h["entry"]) for h in hits])
+        if vectors:
+            hits = collapse_near_duplicate_hits(hits, vectors, k)
+    return hits, engine
 
 
 def search(query: str, limit: int = 10, kind: str | None = None,

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73/boost_skill_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.2.71
+Version: 1.2.73
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License-Expression: Apache-2.0

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73}/boost_skill_cli.egg-info/SOURCES.txt

@@ -210,6 +210,110 @@ docs/roadmap/items/ai-bridge-silent-failure-logging.md
 docs/roadmap/items/ambiguous-tap-short-name-resolution.md
 docs/roadmap/items/atomic-corruption-safe-lock-file-writes.md
 docs/roadmap/items/atomic-skill-install-temp-dir-swap.md
+docs/roadmap/items/audit-absorb-findings.md
+docs/roadmap/items/audit-adapt-findings.md
+docs/roadmap/items/audit-adapt-renders-every-sibling-of-a-flat-agents-dir-workflow-as.md
+docs/roadmap/items/audit-adapt-run-stats-edit-tag-export-reject-the-tap-name-qualifie.md
+docs/roadmap/items/audit-ai-degrade-note-blames-path-api-keys-regardless-of-cause-sev.md
+docs/roadmap/items/audit-attest-findings.md
+docs/roadmap/items/audit-audit-findings.md
+docs/roadmap/items/audit-bmad-findings.md
+docs/roadmap/items/audit-browse-findings.md
+docs/roadmap/items/audit-bundle-findings.md
+docs/roadmap/items/audit-catalog-findings.md
+docs/roadmap/items/audit-catalog-resolve-one-s-duplicate-name-hint-advertises-path-to.md
+docs/roadmap/items/audit-changelog-findings.md
+docs/roadmap/items/audit-chat-findings.md
+docs/roadmap/items/audit-clean-counts-failed-removals-as-cleaned-journals-the-inflate.md
+docs/roadmap/items/audit-cli-py-commands-summaries-and-parser-help-contradict-behavio.md
+docs/roadmap/items/audit-cohort-findings.md
+docs/roadmap/items/audit-completions-findings.md
+docs/roadmap/items/audit-config-findings.md
+docs/roadmap/items/audit-config-policy-set-store-type-unchecked-values-consumers-cras.md
+docs/roadmap/items/audit-context-findings.md
+docs/roadmap/items/audit-corrupt-settings-config-state-json-silently-read-as-empty-th.md
+docs/roadmap/items/audit-create-distill-infer-absorb-install-silently-replace-an-inst.md
+docs/roadmap/items/audit-create-findings.md
+docs/roadmap/items/audit-declined-confirms-never-name-y-boost-assume-yes-snapshot-cle.md
+docs/roadmap/items/audit-deps-findings.md
+docs/roadmap/items/audit-discover-findings.md
+docs/roadmap/items/audit-distill-s-heuristic-merge-drops-repeated-fences-braces-writi.md
+docs/roadmap/items/audit-doctor-findings.md
+docs/roadmap/items/audit-drift-findings.md
+docs/roadmap/items/audit-dry-runs-disagree-with-the-real-run-compact-counts-bytes-it.md
+docs/roadmap/items/audit-edit-findings.md
+docs/roadmap/items/audit-evolve-findings.md
+docs/roadmap/items/audit-explain-findings.md
+docs/roadmap/items/audit-export-findings.md
+docs/roadmap/items/audit-export-o-cohort-create-and-profile-save-silently-overwrite-e.md
+docs/roadmap/items/audit-fingerprint-findings.md
+docs/roadmap/items/audit-five-exit-70-crashes-on-bad-paths-data-catalog-export-serve.md
+docs/roadmap/items/audit-focus-profile-sideline-by-unlinking-without-recording-it-lis.md
+docs/roadmap/items/audit-git-operations-never-set-git-terminal-prompt-0-so-a-404-priv.md
+docs/roadmap/items/audit-health-findings.md
+docs/roadmap/items/audit-hooks-remove-n-cannot-find-hooks-boost-itself-added-unknown.md
+docs/roadmap/items/audit-import-all-skips-the-injection-secret-scans-and-the-per-skil.md
+docs/roadmap/items/audit-import-findings.md
+docs/roadmap/items/audit-index-findings.md
+docs/roadmap/items/audit-info-stats-explain-render-a-different-smaller-shape-for-rule.md
+docs/roadmap/items/audit-install-dry-run-promises-agents-the-real-install-never-write.md
+docs/roadmap/items/audit-install-findings.md
+docs/roadmap/items/audit-install-local-writes-empties-the-repo-s-mcp-json-silently-of.md
+docs/roadmap/items/audit-json-accepted-but-ignored-on-many-branches-cohort-config-pol.md
+docs/roadmap/items/audit-lint-findings.md
+docs/roadmap/items/audit-log-findings.md
+docs/roadmap/items/audit-log-simulate-changelog-headings-echo-the-raw-qualified-argum.md
+docs/roadmap/items/audit-mcp-findings.md
+docs/roadmap/items/audit-mcp-register-s-boost-first-consent-names-only-the-registered.md
+docs/roadmap/items/audit-missing-json-on-doctor-test-health-changelog-trending-log-ho.md
+docs/roadmap/items/audit-name-miss-errors-unknown-tap-qualifier-unnamed-no-close-matc.md
+docs/roadmap/items/audit-onboard-findings.md
+docs/roadmap/items/audit-out-table-clips-data-columns-to-an-assumed-80-columns-when-s.md
+docs/roadmap/items/audit-out-warn-defaults-to-stdout-infer-absorb-corrupt-skill-md-se.md
+docs/roadmap/items/audit-outdated-findings.md
+docs/roadmap/items/audit-output-wrap-tokens-splits-punctuation-off-backtick-spans-ins.md
+docs/roadmap/items/audit-pinned-taps-silently-moved-to-head-by-update-re-clone-and-co.md
+docs/roadmap/items/audit-policy-findings.md
+docs/roadmap/items/audit-preview-findings.md
+docs/roadmap/items/audit-profile-findings.md
+docs/roadmap/items/audit-project-scope-seams-uninstall-verify-list-info-reinstall-dis.md
+docs/roadmap/items/audit-protocol-findings.md
+docs/roadmap/items/audit-pulse-findings.md
+docs/roadmap/items/audit-quarantine-findings.md
+docs/roadmap/items/audit-quarantine-pin-state-invisible-list-info-doctor-report-links.md
+docs/roadmap/items/audit-quickstart-findings.md
+docs/roadmap/items/audit-recommend-findings.md
+docs/roadmap/items/audit-reindex-findings.md
+docs/roadmap/items/audit-reinstall-sync-repair-and-audit-resolve-by-name-tap-ignoring.md
+docs/roadmap/items/audit-replay-findings.md
+docs/roadmap/items/audit-root-findings.md
+docs/roadmap/items/audit-rules-workflows-reported-as-skills-uninstall-claims-a-store.md
+docs/roadmap/items/audit-run-findings.md
+docs/roadmap/items/audit-schedule-findings.md
+docs/roadmap/items/audit-search-findings.md
+docs/roadmap/items/audit-simulate-findings.md
+docs/roadmap/items/audit-singular-plural-misses-1-skills-1-issue-need-1-skill-pass-ac.md
+docs/roadmap/items/audit-six-count-flags-tap-chat-absorb-lint-changelog-hooks-skip-ut.md
+docs/roadmap/items/audit-skill-profile-name-slugging-is-inconsistent-distill-o-accept.md
+docs/roadmap/items/audit-snapshot-restore-replaces-the-lock-wholesale-orphaning-newer.md
+docs/roadmap/items/audit-stale-prose-after-shipped-changes-catalog-export-live-discov.md
+docs/roadmap/items/audit-stray-positionals-and-inapplicable-flags-silently-ignored-ac.md
+docs/roadmap/items/audit-sweep-positionals-json-lack-help-strings-and-no-command-help.md
+docs/roadmap/items/audit-sync-findings.md
+docs/roadmap/items/audit-tag-findings.md
+docs/roadmap/items/audit-tap-dry-run-is-silently-ignored-outside-catalog-spec-and-def.md
+docs/roadmap/items/audit-tap-findings.md
+docs/roadmap/items/audit-taps-findings.md
+docs/roadmap/items/audit-taps-outdated-decay-policy-snapshot-json-emit-display-string.md
+docs/roadmap/items/audit-test-findings.md
+docs/roadmap/items/audit-trust-findings.md
+docs/roadmap/items/audit-trust-verify-labels-a-manifest-tampered-after-signing-by-a-t.md
+docs/roadmap/items/audit-unpin-findings.md
+docs/roadmap/items/audit-untap-findings.md
+docs/roadmap/items/audit-update-findings.md
+docs/roadmap/items/audit-verify-drift-say-nothing-installed-exit-0-and-doctor-says-lo.md
+docs/roadmap/items/audit-verify-findings.md
+docs/roadmap/items/audit-who-findings.md
 docs/roadmap/items/autonomous-ship-workflow-and-isolated-worktree.md
 docs/roadmap/items/bm25-has-no-stemming.md
 docs/roadmap/items/bm25-index-is-one-json-blob.md
@@ -227,6 +331,7 @@ docs/roadmap/items/browse-crashes-selecting-a-rule-or-workflow.md
 docs/roadmap/items/build-provenance-slsa-attestations.md
 docs/roadmap/items/cache-the-catalog-entry-set-across-rag-queries.md
 docs/roadmap/items/capability-manifest-and-least-privilege-policy.md
+docs/roadmap/items/catalog-categories-beyond-curated.md
 docs/roadmap/items/catalogue-entries-carry-no-content-identity.md
 docs/roadmap/items/chat-cites-sources-by-a-number-it-never-prints.md
 docs/roadmap/items/chat-command-grounded-conversational-search.md
@@ -234,6 +339,7 @@ docs/roadmap/items/ci-job-timeout-minutes.md
 docs/roadmap/items/ci-summary-step-fails-itself.md
 docs/roadmap/items/ci-time-is-now-runner-queueing.md
 docs/roadmap/items/clean-env-install-smoke-pip-and-pipx.md
+docs/roadmap/items/cli-audit-2026-08-full-sweep.md
 docs/roadmap/items/cli-command-audit.md
 docs/roadmap/items/cli-output-ignored-the-terminal.md
 docs/roadmap/items/code-scanning-rule-can-be-restored.md
@@ -362,6 +468,7 @@ docs/roadmap/items/mutation-hardening-core-frontmatter-py.md
 docs/roadmap/items/mutation-hardening-core-gitutil-py.md
 docs/roadmap/items/mutation-hardening-core-store-py.md
 docs/roadmap/items/mutation-shard-floor-is-one-file.md
+docs/roadmap/items/mutation-shards-rerun-full-set-every-push.md
 docs/roadmap/items/mypy-strict-mode-for-commands.md
 docs/roadmap/items/near-duplicate-items-eat-the-result-slots.md
 docs/roadmap/items/near-identical-copies-still-eat-the-slots.md
@@ -442,6 +549,7 @@ docs/roadmap/items/self-update-claims-up-to-date-without-asking.md
 docs/roadmap/items/serve-404-reflected-the-request.md
 docs/roadmap/items/serve-is-a-searchable-catalogue.md
 docs/roadmap/items/serve-py-traversal-guards-untested.md
+docs/roadmap/items/shard-refresh-skips-processed-commits.md
 docs/roadmap/items/shard-runs-reembed-everything-weekly.md
 docs/roadmap/items/shards-pipeline-never-produced-a-shard.md
 docs/roadmap/items/shareable-catalogue-bundle.md
@@ -695,6 +803,7 @@ tests/unit/test_dense.py
 tests/unit/test_dense_boilerplate_cluster.py
 tests/unit/test_dense_embed_dedupe.py
 tests/unit/test_dense_entry_reuse.py
+tests/unit/test_dense_entry_vectors.py
 tests/unit/test_dense_fallback.py
 tests/unit/test_dense_fix_hint.py
 tests/unit/test_dense_progress.py
@@ -761,6 +870,7 @@ tests/unit/test_mutation_hardening.py
 tests/unit/test_mutation_shards.py
 tests/unit/test_mutation_subfile_shards.py
 tests/unit/test_mutmut_source_paths.py
+tests/unit/test_near_duplicate_collapse.py
 tests/unit/test_nethttp.py
 tests/unit/test_normalize_sdist.py
 tests/unit/test_output.py
@@ -790,6 +900,7 @@ tests/unit/test_release_preflight.py
 tests/unit/test_required_checks.py
 tests/unit/test_rerank_cache.py
 tests/unit/test_resolve.py
+tests/unit/test_retrieve_any_near_duplicates.py
 tests/unit/test_roadmap_fresh.py
 tests/unit/test_rules.py
 tests/unit/test_scale_corpus.py

{boost_skill_cli-1.2.71 → boost_skill_cli-1.2.73}/docs/commands.html

@@ -117,7 +117,7 @@
     </div>
     <div class="group" data-group="find">
       <div class="group-h">Find &amp; search</div>
-      <a class="cmd-link" href="#cmd-search" data-search="search search skills across tap registries (ai-ranked)  --smart --limit limit --json query">search</a>
+      <a class="cmd-link" href="#cmd-search" data-search="search search skills across tap registries (ai-ranked)  --smart --limit limit --collapse-near-duplicates --json query">search</a>
       <a class="cmd-link" href="#cmd-reindex" data-search="reindex build/refresh the full-content search index build or refresh the full-content search index --force --dense --export-shard tap --import-shard file --fetch-shards --json">reindex</a>
       <a class="cmd-link" href="#cmd-discover" data-search="discover search github for skill repos you have not tapped yet search github for skill repositories you have not tapped yet --limit limit --json --local query">discover</a>
       <a class="cmd-link" href="#cmd-recommend" data-search="recommend suggest skills based on your project&#x27;s tech stack  --path path --limit limit --json">recommend</a>
@@ -397,10 +397,10 @@
         </div>
       </section>
       <h2 class="group-h" id="grp-find">Find &amp; search</h2>
-      <section class="cmd" id="cmd-search" data-search="search search skills across tap registries (ai-ranked)  --smart --limit limit --json query">
+      <section class="cmd" id="cmd-search" data-search="search search skills across tap registries (ai-ranked)  --smart --limit limit --collapse-near-duplicates --json query">
         <h3><span class="cname">search</span><span class="gtag">find</span></h3>
         <p class="summary">Search skills across tap registries (AI-ranked)</p>
-        <pre class="syn">boost search [--smart] [--limit LIMIT] [--json] query [query ...]</pre>
+        <pre class="syn">boost search [--smart] [--limit LIMIT] [--collapse-near-duplicates] [--json] query [query ...]</pre>
         <div class="args-h">Arguments</div>
         <div class="args">
           <div class="arg"><code>query</code><span>search terms</span></div>
@@ -409,6 +409,7 @@
         <div class="args">
           <div class="arg"><code>--smart</code><span>rerank the top hits with Claude</span></div>
           <div class="arg"><code>--limit LIMIT</code><span>max results (default 15)</span></div>
+          <div class="arg"><code>--collapse-near-duplicates</code><span>collapse near-identical entries (translations, paraphrases) via dense embeddings — opt-in, needs a built dense index; threshold not yet validated against a real corpus, see the roadmap</span></div>
           <div class="arg"><code>--json</code><span>machine-readable output</span></div>
         </div>
       </section>

boost_skill_cli-1.2.73/docs/roadmap/items/audit-absorb-findings.md

@@ -0,0 +1,35 @@
+---
+id: audit-absorb-findings
+board: code
+section: dx
+status: planned
+category: CLI · Consistency
+complexity: S
+impact: Low
+wow: 1
+note: absorb is the only generated-skill command with no journal.log call at all
+order: 246
+owner:
+pr:
+title: "boost absorb: CLI audit findings (2026-08)"
+---
+<b><code>cmd_absorb</code> never calls <code>journal.log</code>.</b> After two
+<code>absorb --install</code> runs, <code>boost log</code> lists <em>&ldquo;infer
+project-conventions&rdquo;</em>, <em>&ldquo;distill brainstorming-distilled&rdquo;</em> and an
+<em>&ldquo;import absorbed-patterns&rdquo;</em> entry &mdash; the last written by
+<code>install_from_path</code> (<code>core/store.py:1228</code>), not by absorb &mdash; and the verb
+<em>absorb</em> never appears. Without <code>--install</code> (stdout mode) the command leaves
+<b>zero</b> journal trace, while its siblings journal in both output branches: distill at
+<code>commands/intelligence.py:187</code>, infer at <code>:326</code>/<code>:330</code>. An omission,
+not a choice &mdash; absorb is the only generated-skill command with no <code>journal.log</code> call.
+
+<br><br>The auditor's second half &mdash; that <code>--install</code> output differs from
+<code>install</code>'s report &mdash; reproduced but was <b>declined on design review</b>:
+<code>_install_generated</code> (<code>intelligence.py:103-125</code>) is the deliberate shared
+renderer for all generated skills (distill/infer/absorb), and reusing <code>store.install</code>'s tap
+report would misreport a local re-import as a fresh tap install. Fix is the journal line alone: add
+<code>journal.log("absorb", name, patterns=len(patterns), files=len(files))</code> after generation in
+both branches of <code>cmd_absorb</code> (<code>intelligence.py:528-580</code>), mirroring distill's
+placement. No doc changes.
+
+<br><br>Found by the 2026-08 CLI audit (cluster <code>absorb-parity</code>); repro in the audit log.

boost_skill_cli-1.2.73/docs/roadmap/items/audit-adapt-findings.md

@@ -0,0 +1,59 @@
+---
+id: audit-adapt-findings
+board: code
+section: dx
+status: planned
+category: CLI · Bug
+complexity: M
+impact: Med
+wow: 2
+note: a subagent named after a tool renders a crew that compiles but cannot run
+order: 247
+owner:
+pr:
+title: "boost adapt: CLI audit findings (2026-08)"
+---
+<b>A subagent named after a declared tool renders modules that compile but cannot run.</b> With a
+subagent <code>grep</code> and tool <code>Grep</code>: crewai emits <code>@tool("grep") def
+grep</code> then <code>grep = Agent(...)</code>, so <code>reviewer_1 = Agent(tools=[read,
+grep])</code> hands the Agent where a tool belongs (stub run: TypeError); langgraph assigns
+<code>grep = create_react_agent(...)</code> inside <code>build_mycrew</code>, making it local, so the
+earlier <code>tools=[read, grep]</code> raises UnboundLocalError. <code>_unique_idents</code>
+(<code>core/adapters.py:188-202</code>) dedups only among agent specs and <code>_unique_tools</code>
+(<code>:205-212</code>) allocates stub names independently. Fix: pass the tool ident set into
+<code>_unique_idents</code> as pre-reserved names (or prefix stubs <code>tool_&lt;name&gt;</code>) in
+<code>render_crew</code>/<code>render_graph</code>, plus a golden test that executes a colliding
+render against stubs.
+
+<br><br><b><code>docs/commands.html</code> brackets required options as optional.</b> Line 369 shows
+<code>boost adapt [--to FRAMEWORK] &hellip;</code> while <code>adapt --help</code> prints an
+unbracketed <code>--to FRAMEWORK</code> and omitting it exits 2 &mdash; and the verify pass found it
+is broader than adapt: evolve's required <code>--feedback</code> and catalog's required
+mutually-exclusive group render all-optional too. Pure generator bug:
+<code>scripts/build_command_reference.py:122-126</code> brackets every option unconditionally. Emit
+required options unbracketed (a required group as <code>(--a | --b)</code>), prefer the short flag
+like argparse, then <code>make generate</code> &mdash; the <code>--check</code> gate holds it after
+that.
+
+<br><br><b>Colon-form model ids get double-prefixed for the LiteLLM targets.</b>
+<code>--model anthropic:claude-x</code> &mdash; the form langgraph accepts <em>and emits</em> &mdash;
+renders <code>llm=LLM(model="anthropic/anthropic:claude-x")</code> for crewai and the same for
+agents-sdk; multi-agent crews inherit it via <code>adapters.py:376</code>.
+<code>_litellm_model</code>'s docstring says a provider-qualified value passes through, but the code
+checks only <code>/</code>. Fix: replace the first <code>:</code> with <code>/</code> before deciding
+to prefix (mirror of <code>_langchain_model</code>); document accepted syntaxes in
+<code>docs/adapters.html</code>'s <code>--model</code> paragraph (~line 344, slash form only today),
+and regenerate <code>docs/commands.html</code> only if the argparse help changes.
+
+<br><br><b><code>adapt -o</code> and <code>run --print -o</code> write generated source mode
+0600</b> &mdash; and re-rendering over an existing 0644 file silently downgrades it, unlike a shell
+redirect (<code>-rw-------</code> vs <code>-rw-r--r--</code> under umask 022).
+<code>util.atomic_write_text</code> (<code>core/util.py:91-116</code>) inherits mkstemp's 0600, right
+for the lock/config it was written for, wrong for source the user asked boost to write. Fix: add an
+optional <code>mode</code> parameter (fchmod the temp fd before <code>os.replace</code>), keep 0600
+the default, and have <code>cmd_adapt</code> (<code>pkg.py:1753-1761</code>) and <code>cmd_run</code>
+(<code>run.py:62</code>) pass the umask default.
+
+<br><br>Found by the 2026-08 CLI audit (clusters <code>adapt-ident-collision</code>,
+<code>docs-required-flag-synopsis</code>, <code>adapt-model-id-syntax</code>,
+<code>generated-file-mode</code>); repro in the audit log.

boost_skill_cli-1.2.73/docs/roadmap/items/audit-adapt-renders-every-sibling-of-a-flat-agents-dir-workflow-as.md

@@ -0,0 +1,22 @@
+---
+id: audit-adapt-renders-every-sibling-of-a-flat-agents-dir-workflow-as
+board: code
+section: internals
+status: planned
+category: CLI · Bug
+complexity: S
+impact: High
+wow: 2
+note: one flat agents/x.md adapts into a 383 KB crew of 138 Agent()s — including itself, twice
+order: 200
+owner:
+pr:
+title: "<code>adapt</code> renders every sibling of a flat <code>agents/</code>-dir workflow as a 138-agent crew"
+---
+<code>adapt actix-expert --to crewai -o x.py</code> prints <em>&ldquo;&#10003; adapted actix-expert &rarr; x.py (crew of 138 &middot; claude-haiku-4-5-20251001)&rdquo;</em> and writes a <b>383 KB</b> render containing <b>138 <code>Agent()</code> + 138 <code>Task()</code></b> &mdash; <code>actix_expert</code>, then <code>actix_expert_1</code> (the item itself, again), then android-expert, angular-expert, &hellip; every sibling in the registry's <code>agents/</code> directory. <code>--to agents-sdk</code> announces <em>&ldquo;note: actix-expert declares 137 subagent(s)&rdquo;</em>. The item is one flat frontmatter file declaring nothing.
+
+The mechanism: <code>cmd_adapt</code> (<code>boost_cli/commands/pkg.py:1737</code>) passes <code>skill_md.parent</code> to <code>discover_subagents</code> (<code>boost_cli/core/adapters.py:157-186</code>). For a SKILL.md-rooted skill that parent is the skill's own directory and the feature works as shipped (<code>adapt brainstorming --to crewai</code> renders a single agent). For a flat <code>agents/&lt;x&gt;.md</code> workflow the parent <em>is</em> the registry-wide <code>agents/</code> dir, so all 137 siblings become a fabricated crew &mdash; and since the exclusion at <code>adapters.py:170-173</code> skips only <code>SKILL.md</code>, the adapted item is re-included as one of its own subagents.
+
+Fix per the verified recommendation: run <code>discover_subagents</code> only when the resolved item is SKILL.md-rooted (<code>skill_md.name == 'SKILL.md'</code>), and harden <code>discover_subagents</code> to skip the entry's own file and require the <code>agents/</code>/<code>subagents/</code> dir to be strictly beneath <code>skill_dir</code>. Add a unit test adapting a bare <code>agents/&lt;x&gt;.md</code> item that asserts a single-<code>Agent()</code> render. Update <code>docs/adapters.html</code> (subagent-discovery wording).
+
+Found by the 2026-08 CLI audit (cluster <code>adapt-subagent-misdetect</code>); repro in the audit log. Verified 2026-08-31: reproduced, defect in the shipped multi-agent adapter feature (whose design intended skill-declared subagents only), not a duplicate of it.