boost-skill-cli 1.2.19__tar.gz → 1.2.20__tar.gz

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.2.19
+Version: 1.2.20
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License: GNU GENERAL PUBLIC LICENSE

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/boost_cli/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.2.19'
-__version_tuple__ = version_tuple = (1, 2, 19)
+__version__ = version = '1.2.20'
+__version_tuple__ = version_tuple = (1, 2, 20)
 
 __commit_id__ = commit_id = None

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/boost_cli/cli.py

@@ -155,20 +155,50 @@ _GROUP_HUES = ("cyan", "violet", "pink", "green", "yellow")
 
 
 def print_help() -> None:
-    print(out.gradient(PRODUCT) + " — Homebrew for AI coding skills"
-          + out.c("  (%s · v%s)" % (TAGLINE, __version__), out.DIM))
+    """The command index, fitted to the pane.
+
+    Every other screen in the CLI adapts — `search` drops the tap column and
+    truncates at 60 columns — while this one emitted a fixed 105-column banner
+    at every width and wrapped mid-word for anyone not running a wide
+    terminal. Names are never clipped, because the help's whole job is to be
+    an index you can copy a command out of; summaries are.
+    """
+    cols = out.term_width()
+
+    title_line = out.gradient(PRODUCT) + " — Homebrew for AI coding skills"
+    stamp = "%s · v%s" % (TAGLINE, __version__)
+    if out.visible_len(title_line) + out.visible_len(stamp) + 4 <= cols:
+        print(title_line + out.c("  (%s)" % stamp, out.DIM))
+    else:
+        print(title_line)
+        # The tagline is the first thing to go; the version never is, because
+        # it is what a bug report needs.
+        print(out.c(stamp if out.visible_len(stamp) <= cols
+                    else "v%s" % __version__, out.DIM))
     print()
-    print(out.c("Usage:", out.BOLD) + "  boost <command> [args]   "
-          + out.c("boost help <command> for details", out.DIM))
+
+    usage = out.c("Usage:", out.BOLD) + "  boost <command> [args]"
+    detail = out.c("boost help <command> for details", out.DIM)
+    if out.visible_len(usage) + out.visible_len(detail) + 3 <= cols:
+        print(usage + "   " + detail)
+    else:
+        print(usage)
+        print("  " + detail)
+
     width = max(len(n) for n, _, _, _ in COMMANDS)
     for idx, (gkey, (_icon, title, desc)) in enumerate(GROUPS.items()):
         cmds = [(n, s) for n, g, _m, s in COMMANDS if g == gkey]
         bullet = out.aurora("●", _GROUP_HUES[idx % len(_GROUP_HUES)])
         print()
-        print(bullet + " " + out.c(title, out.BOLD)
-              + out.c("  — %s" % desc, out.DIM))
+        head = bullet + " " + out.c(title, out.BOLD)
+        room = cols - len(title) - 6  # "● " + title + "  — "
+        if room >= 8:
+            print(head + out.c("  — %s" % out.truncate(desc, room), out.DIM))
+        else:
+            print(head)
         for n, s in cmds:
-            print("  " + out.c(n.ljust(width + 2), out.CYAN) + s)
+            print("  " + out.c(n.ljust(width + 2), out.CYAN)
+                  + out.truncate(s, max(0, cols - width - 4)))
     print()
     print(out.c("%d commands · %d groups" % (len(COMMANDS), len(GROUPS)), out.DIM))
 

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/boost_cli/commands/team.py

@@ -69,7 +69,7 @@ def cmd_cohort(argv) -> int:
     p = cliparse.parser(
         prog="boost cohort",
         description="Controlled skill rollouts & team A/B testing",
-        epilog="Membership is a deterministic hash of user+cohort, so a 50%% "
+        epilog="Membership is a deterministic hash of user+cohort, so a 50% "
                "rollout lands on the same half of the team every time. "
                "This machine evaluates its own membership locally.")
     p.add_argument("action", nargs="?", default="list",
@@ -177,7 +177,7 @@ def cmd_cohort(argv) -> int:
         return 0
     out.table(rows, headers=("COHORT", "SKILLS", "ROLLOUT", "YOU"))
     print()
-    out.dim("membership = sha256(user:cohort) %% 100 < rollout · apply with `boost cohort apply`")
+    out.dim("membership = sha256(user:cohort) % 100 < rollout · apply with `boost cohort apply`")
     return 0
 
 

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/boost_cli/core/output.py

@@ -317,9 +317,20 @@ def panel(lines, title: str | None = None, hue: str = "cyan") -> str:
     """
     if isinstance(lines, str):
         lines = [lines]
+    # The border costs four columns — "\u2502 " on the left, " \u2502" on the right —
+    # so that is what the content may not exceed. Without this the box sized
+    # itself to its content and sailed past the pane: `boost count` drew 108
+    # columns into an 80-column terminal, and a box whose border wraps is the
+    # worst-looking overflow the CLI has, because the shape itself breaks.
+    room = term_width() - 4
+    if title:
+        title = _clip_visible(title, room - 2)
+    lines = [_clip_visible(x, room) for x in lines]
     widths = [visible_len(x) for x in lines]
     tw = visible_len(title) if title else 0
-    inner = max([*widths, tw + 2])  # a titled rule needs a space each side
+    # A titled rule needs a space each side, hence the +2 / -2. Clipping above
+    # already bounds every width by `room`, so no second clamp is needed here.
+    inner = max([*widths, tw + 2])
 
     def b(s: str) -> str:
         return aurora(s, hue)

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/boost_skill_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.2.19
+Version: 1.2.20
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License: GNU GENERAL PUBLIC LICENSE

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/boost_skill_cli.egg-info/SOURCES.txt

@@ -191,6 +191,7 @@ 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/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
 docs/roadmap/items/bmad-autopilot-one-command-persona-routing.md
 docs/roadmap/items/boost-first-carried-the-losing-trigger.md
@@ -211,6 +212,7 @@ 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-output-ignored-the-terminal.md
 docs/roadmap/items/code-scanning-rule-can-be-restored.md
 docs/roadmap/items/codeql-job-rename-stranded-merge-protection.md
 docs/roadmap/items/cold-search-reads-the-whole-catalogue.md
@@ -296,6 +298,7 @@ docs/roadmap/items/localize-the-stored-bm25-snippet.md
 docs/roadmap/items/lock-invariant-cannot-parse-extras-pins.md
 docs/roadmap/items/lockfile-enforcement-and-commit-pinning.md
 docs/roadmap/items/log-timestamps-mislabeled-utc.md
+docs/roadmap/items/long-hints-overflow-narrow-panes.md
 docs/roadmap/items/lowest-version-resolution-uv-resolution-lowest-d.md
 docs/roadmap/items/main-has-no-branch-protection.md
 docs/roadmap/items/main-ruleset-ref-pattern-has-literal-quotes.md

boost_skill_cli-1.2.20/docs/roadmap/items/bm25-has-no-stemming.md

@@ -0,0 +1,50 @@
+---
+id: bm25-has-no-stemming
+board: code
+section: pipeline
+status: next
+category: Search · Retrieval
+complexity: M
+impact: High
+wow: 4
+note: boost search brainstorm finds nothing while brainstorming finds it; 15.7% of catalog names are unreachable by their own stem
+order: 127
+owner:
+pr:
+title: <code>boost search brainstorm</code> finds nothing, and <code>brainstorming</code> finds it
+---
+<b>The symptom, in two commands against the same five-skill tap.</b>
+<code>boost search brainstorming</code> returns the skill.
+<code>boost search brainstorm</code> returns <b>zero</b> matches and sends the user to
+<code>boost discover</code> to search all of GitHub for something already sitting in their
+catalogue. <code>boost chat "what helps me brainstorm"</code> has the same hole and answers with
+two unrelated skills. <code>core/rag.tokenize</code> lowercases, splits on non-alphanumerics and
+drops a small stopword list — there is no stemming and no prefix fallback, so
+<code>brainstorm</code> and <code>brainstorming</code> are simply different terms.
+
+<b>Measured on the 461 tap caches on this machine: 29,607 distinct item names, 9,739 distinct
+name tokens.</b> <b>4,663 of those names (15.7%)</b> carry a token whose stem finds nothing —
+counting only stems that appear at least five times elsewhere in the corpus, so each is
+demonstrably a word people type rather than an artefact of naive suffix stripping. The worst are
+not exotic: <code>pattern</code> misses <b>474</b> names ending <code>-patterns</code>,
+<code>skill</code> misses <b>252</b> ending <code>-skills</code>, <code>implement</code> misses
+161, <code>event</code> 115, <code>error</code> 112, <code>webhook</code> 109,
+<code>agent</code> 83, <code>market</code> 81 named <code>-marketing</code>. A user searching
+this catalogue for <code>skill</code> cannot reach a quarter of a thousand items named for it.
+
+<b>It is invisible from a developer install, which is why it survived.</b> This machine has the
+<code>[rag]</code> extra and a built dense store, so search reports <code>hybrid RRF (BM25 +
+dense)</code> and the embeddings absorb the morphology: <code>webhook</code> returns 51 matches
+and <code>webhooks</code> returns 47. The gap is only visible on the always-on, zero-dependency
+BM25 path — which is what every user without the extra runs, what
+<code>search</code> prints <code>semantic search is off</code> for, and what the required
+<code>eval</code> gate floors. There is no <code>--engine</code> flag, so a maintainer cannot
+easily reproduce a plain user's result.
+
+<b>Two fixes, and they are not the same size.</b> A <b>zero-match prefix fallback</b> — on 0
+hits, retry against vocabulary tokens by prefix — touches no index format, needs no baseline
+regeneration, and fires only where BM25 currently returns nothing, so it cannot move a single
+eval metric; it fixes the exact observed symptom in both <code>search</code> and
+<code>chat</code>. A <b>real stemmer</b> is the better answer and costs an index-format bump plus
+a regenerated <code>tests/eval/baseline.json</code>, and would move all four floored metrics.
+Either lands in <code>core/rag.py</code>, so the tests must kill mutants, not merely cover lines.

boost_skill_cli-1.2.20/docs/roadmap/items/cli-output-ignored-the-terminal.md

@@ -0,0 +1,56 @@
+---
+id: cli-output-ignored-the-terminal
+board: code
+section: dx
+status: inflight
+category: CLI · Output
+complexity: S
+impact: Med
+wow: 3
+note: an 80-command sweep found a box that drew wider than the pane, a help screen that never measured it, and a literal %% on screen
+order: 125
+owner: loop/cli-output-defects
+pr: 552
+title: The box drew 108 columns into an 80-column pane, and <code>--help</code> never asked how wide the pane was
+---
+<b>Every one of the 80 commands was run, not just read.</b> All 80 answered
+<code>--help</code> with exit 0 and no traceback; ~73 were then driven for their real effect
+against a disposable <code>HOME</code> — install, link, quarantine, focus, snapshot, replay,
+cohort, hooks and the rest, checking the filesystem after each rather than trusting the success
+line. <code>boost focus tdd-workflow</code> was confirmed by listing
+<code>~/.claude/skills</code> and finding the other two skills genuinely unlinked with the
+canonical store intact, and <code>focus --clear</code> by finding all three back. Seven commands
+could not run here and are recorded as blocked rather than passed: <code>serve</code> (the
+sandbox denies <code>socket.bind</code>), <code>discover</code> and <code>index</code> (no
+<code>gh</code> auth), <code>run</code> live (no Agents SDK — <code>--print</code> was verified
+instead), plus the interactive halves of <code>browse</code>, <code>chat</code> and
+<code>edit</code>, each of which degraded with a correct hint.
+
+<b>A literal <code>%%</code> was reaching the terminal.</b>
+<code>boost cohort</code> printed <code>sha256(user:cohort) %% 100 &lt; rollout</code> and its
+<code>--help</code> epilog offered <code>a 50%% rollout</code>. Both strings carry printf
+escaping and neither is ever <code>%</code>-formatted, so the escape survived to the screen. The
+existing test asserted <code>"membership = sha256(user:cohort)" in r.out</code> — it stops one
+character short of the defect, which is exactly how it survived. Four other <code>%%</code> in
+the same file are correct: they sit inside real <code>%</code>-format calls, including the
+<code>Exec=%s %%u</code> that a <code>.desktop</code> file requires.
+
+<b><code>panel()</code> sized itself to its content and never asked the terminal.</b> Measured:
+<code>boost count</code> drew <b>108 columns into an 80-column pane</b>. A box is the worst
+thing to overflow, because the border wraps and the shape itself breaks. It now clamps to
+<code>term_width() - 4</code> — the four columns the border costs — and clips over-long content,
+so every row stays the same width at any pane size.
+
+<b>The help screen was the one screen that never measured the pane.</b>
+<code>search</code> already adapts, dropping the tap column and truncating at 60 columns —
+<code>term_width()</code> existed and <code>discovery.py</code> was its only caller. Meanwhile
+<code>boost --help</code>, the first thing a new user sees, emitted a fixed <b>102-column</b>
+banner at every width. It now fits exactly at 60, 80 and 100. Command <em>names</em> are never
+clipped, because the help's whole job is to be an index you can copy a command out of; summaries
+and the tagline are. The version never is — it is what a bug report needs.
+
+<b>Two findings were measured and deliberately left for their own items.</b> Eleven commands
+still overflow 80 columns, but the remainder are prose hints, and the worst of them is one shared
+string that six test files and a BDD feature pin by substring — wrapping it is a cross-cutting
+output change, not a rider on this one. See <code>long-hints-overflow-narrow-panes</code> and
+<code>bm25-has-no-stemming</code>.

boost_skill_cli-1.2.20/docs/roadmap/items/long-hints-overflow-narrow-panes.md

@@ -0,0 +1,38 @@
+---
+id: long-hints-overflow-narrow-panes
+board: code
+section: dx
+status: next
+category: CLI · Output
+complexity: M
+impact: Low
+wow: 2
+note: 11 commands still emit prose hints past 80 columns; the worst is one string pinned by six test files
+order: 126
+owner:
+pr:
+title: The hints still run past the pane, and the worst one is pinned by six test files
+---
+<b>Measured, at <code>COLUMNS=80</code>, over 46 commands.</b> After the box and the help screen
+were fitted, <b>11</b> still emit a line past 80 columns. Ranked:
+<code>doctor</code>'s semantic-search hint at <b>127</b>; the AI-fallback notice at <b>101</b>,
+which is one shared string surfacing in <code>explain</code>, <code>impact</code> and
+<code>simulate</code>; <code>protocol</code>'s URL forms at 100; <code>changelog</code>'s
+unshallow hint at 93; <code>who</code>'s footer at 87; <code>search</code>'s extra hint at 82;
+<code>replay</code>'s footer at 81.
+
+<b>Two of the eleven must be left alone.</b> <code>pulse</code>'s <code>source=</code> paths and
+<code>fingerprint</code>'s hash are data, not chrome — clipping them destroys the information the
+line exists to carry. So a blanket "no line exceeds <code>term_width()</code>" assertion is the
+wrong gate: it would false-positive on exactly the lines that should be long. Pin the composed
+sites individually.
+
+<b>Why this is not a quick wrap.</b> The 101-column offender contains
+<code>using the heuristic fallback</code>, which is asserted by <b>six</b> test files plus
+<code>tests/bdd/features/explain.feature</code>. Wrapping inserts a newline mid-sentence and
+breaks every one of them; and under pytest <code>term_width()</code> falls back to 80, so the
+wrap fires in the suite rather than only on a narrow terminal. The change is a
+<code>wrap()</code> helper in <code>core/output.py</code> — mutation-gated, so boundary-width
+tests, not coverage — plus re-pinning those assertions. Shortening the strings instead is a
+smaller diff but a voice decision, and the substring pinned by the tests would have to survive
+either way.

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/docs/roadmap.html

@@ -133,7 +133,7 @@
       <div class="stat"><b>16</b><span>Shipped</span></div>
       <div class="stat"><b>1</b><span>Next up</span></div>
       <div class="stat"><b>7</b><span>Planned</span></div>
-      <div class="stat"><b>264</b><span>Loop finds</span></div>
+      <div class="stat"><b>267</b><span>Loop finds</span></div>
       <!-- /ROADMAP:stats -->
     </div>
   </section>
@@ -7053,6 +7053,52 @@ state where that line is the advice.</p>
           <span class="m">0.94 s cold search at 71.6k entries — ~0.5 s spent materialising data the top hits never use</span>
         </div>
       </article>
+
+      <article class="cap rcard" id="bm25-has-no-stemming">
+        <div class="head"><span class="pill next">Next</span><span class="cat">Search · Retrieval</span></div>
+        <h3><code>boost search brainstorm</code> finds nothing, and <code>brainstorming</code> finds it</h3>
+        <p><b>The symptom, in two commands against the same five-skill tap.</b>
+<code>boost search brainstorming</code> returns the skill.
+<code>boost search brainstorm</code> returns <b>zero</b> matches and sends the user to
+<code>boost discover</code> to search all of GitHub for something already sitting in their
+catalogue. <code>boost chat "what helps me brainstorm"</code> has the same hole and answers with
+two unrelated skills. <code>core/rag.tokenize</code> lowercases, splits on non-alphanumerics and
+drops a small stopword list — there is no stemming and no prefix fallback, so
+<code>brainstorm</code> and <code>brainstorming</code> are simply different terms.
+
+<b>Measured on the 461 tap caches on this machine: 29,607 distinct item names, 9,739 distinct
+name tokens.</b> <b>4,663 of those names (15.7%)</b> carry a token whose stem finds nothing —
+counting only stems that appear at least five times elsewhere in the corpus, so each is
+demonstrably a word people type rather than an artefact of naive suffix stripping. The worst are
+not exotic: <code>pattern</code> misses <b>474</b> names ending <code>-patterns</code>,
+<code>skill</code> misses <b>252</b> ending <code>-skills</code>, <code>implement</code> misses
+161, <code>event</code> 115, <code>error</code> 112, <code>webhook</code> 109,
+<code>agent</code> 83, <code>market</code> 81 named <code>-marketing</code>. A user searching
+this catalogue for <code>skill</code> cannot reach a quarter of a thousand items named for it.
+
+<b>It is invisible from a developer install, which is why it survived.</b> This machine has the
+<code>[rag]</code> extra and a built dense store, so search reports <code>hybrid RRF (BM25 +
+dense)</code> and the embeddings absorb the morphology: <code>webhook</code> returns 51 matches
+and <code>webhooks</code> returns 47. The gap is only visible on the always-on, zero-dependency
+BM25 path — which is what every user without the extra runs, what
+<code>search</code> prints <code>semantic search is off</code> for, and what the required
+<code>eval</code> gate floors. There is no <code>--engine</code> flag, so a maintainer cannot
+easily reproduce a plain user's result.
+
+<b>Two fixes, and they are not the same size.</b> A <b>zero-match prefix fallback</b> — on 0
+hits, retry against vocabulary tokens by prefix — touches no index format, needs no baseline
+regeneration, and fires only where BM25 currently returns nothing, so it cannot move a single
+eval metric; it fixes the exact observed symptom in both <code>search</code> and
+<code>chat</code>. A <b>real stemmer</b> is the better answer and costs an index-format bump plus
+a regenerated <code>tests/eval/baseline.json</code>, and would move all four floored metrics.
+Either lands in <code>core/rag.py</code>, so the tests must kill mutants, not merely cover lines.</p>
+        <div class="meta">
+          <span class="m">Complexity <b>M</b></span>
+          <span class="m hi">Impact <b>High</b></span>
+          <span class="m">Wow <b class="wow">★★★★</b></span>
+          <span class="m">boost search brainstorm finds nothing while brainstorming finds it; 15.7% of catalog names are unreachable by their own stem</span>
+        </div>
+      </article>
       <!-- /ROADMAP:cards -->
 
     </div>
@@ -9009,6 +9055,92 @@ the same search from 0.94 s to 0.49 s.</p>
           <span class="m">MCP boost_search measured 11.7-17 s per call — every call, even a repeat of the last one</span>
         </div>
       </article>
+
+      <article class="cap rcard" id="cli-output-ignored-the-terminal">
+        <div class="head"><span class="pill inflight">In flight</span><span class="cat">CLI · Output</span></div>
+        <h3>The box drew 108 columns into an 80-column pane, and <code>--help</code> never asked how wide the pane was</h3>
+        <p><b>Every one of the 80 commands was run, not just read.</b> All 80 answered
+<code>--help</code> with exit 0 and no traceback; ~73 were then driven for their real effect
+against a disposable <code>HOME</code> — install, link, quarantine, focus, snapshot, replay,
+cohort, hooks and the rest, checking the filesystem after each rather than trusting the success
+line. <code>boost focus tdd-workflow</code> was confirmed by listing
+<code>~/.claude/skills</code> and finding the other two skills genuinely unlinked with the
+canonical store intact, and <code>focus --clear</code> by finding all three back. Seven commands
+could not run here and are recorded as blocked rather than passed: <code>serve</code> (the
+sandbox denies <code>socket.bind</code>), <code>discover</code> and <code>index</code> (no
+<code>gh</code> auth), <code>run</code> live (no Agents SDK — <code>--print</code> was verified
+instead), plus the interactive halves of <code>browse</code>, <code>chat</code> and
+<code>edit</code>, each of which degraded with a correct hint.
+
+<b>A literal <code>%%</code> was reaching the terminal.</b>
+<code>boost cohort</code> printed <code>sha256(user:cohort) %% 100 &lt; rollout</code> and its
+<code>--help</code> epilog offered <code>a 50%% rollout</code>. Both strings carry printf
+escaping and neither is ever <code>%</code>-formatted, so the escape survived to the screen. The
+existing test asserted <code>"membership = sha256(user:cohort)" in r.out</code> — it stops one
+character short of the defect, which is exactly how it survived. Four other <code>%%</code> in
+the same file are correct: they sit inside real <code>%</code>-format calls, including the
+<code>Exec=%s %%u</code> that a <code>.desktop</code> file requires.
+
+<b><code>panel()</code> sized itself to its content and never asked the terminal.</b> Measured:
+<code>boost count</code> drew <b>108 columns into an 80-column pane</b>. A box is the worst
+thing to overflow, because the border wraps and the shape itself breaks. It now clamps to
+<code>term_width() - 4</code> — the four columns the border costs — and clips over-long content,
+so every row stays the same width at any pane size.
+
+<b>The help screen was the one screen that never measured the pane.</b>
+<code>search</code> already adapts, dropping the tap column and truncating at 60 columns —
+<code>term_width()</code> existed and <code>discovery.py</code> was its only caller. Meanwhile
+<code>boost --help</code>, the first thing a new user sees, emitted a fixed <b>102-column</b>
+banner at every width. It now fits exactly at 60, 80 and 100. Command <em>names</em> are never
+clipped, because the help's whole job is to be an index you can copy a command out of; summaries
+and the tagline are. The version never is — it is what a bug report needs.
+
+<b>Two findings were measured and deliberately left for their own items.</b> Eleven commands
+still overflow 80 columns, but the remainder are prose hints, and the worst of them is one shared
+string that six test files and a BDD feature pin by substring — wrapping it is a cross-cutting
+output change, not a rider on this one. See <code>long-hints-overflow-narrow-panes</code> and
+<code>bm25-has-no-stemming</code>.</p>
+        <div class="meta">
+          <span class="m">Complexity <b>S</b></span>
+          <span class="m">Impact <b>Med</b></span>
+          <span class="m">Wow <b class="wow">★★★</b></span>
+          <span class="m">an 80-command sweep found a box that drew wider than the pane, a help screen that never measured it, and a literal %% on screen</span>
+        </div>
+      </article>
+
+      <article class="cap rcard" id="long-hints-overflow-narrow-panes">
+        <div class="head"><span class="pill next">Next</span><span class="cat">CLI · Output</span></div>
+        <h3>The hints still run past the pane, and the worst one is pinned by six test files</h3>
+        <p><b>Measured, at <code>COLUMNS=80</code>, over 46 commands.</b> After the box and the help screen
+were fitted, <b>11</b> still emit a line past 80 columns. Ranked:
+<code>doctor</code>'s semantic-search hint at <b>127</b>; the AI-fallback notice at <b>101</b>,
+which is one shared string surfacing in <code>explain</code>, <code>impact</code> and
+<code>simulate</code>; <code>protocol</code>'s URL forms at 100; <code>changelog</code>'s
+unshallow hint at 93; <code>who</code>'s footer at 87; <code>search</code>'s extra hint at 82;
+<code>replay</code>'s footer at 81.
+
+<b>Two of the eleven must be left alone.</b> <code>pulse</code>'s <code>source=</code> paths and
+<code>fingerprint</code>'s hash are data, not chrome — clipping them destroys the information the
+line exists to carry. So a blanket "no line exceeds <code>term_width()</code>" assertion is the
+wrong gate: it would false-positive on exactly the lines that should be long. Pin the composed
+sites individually.
+
+<b>Why this is not a quick wrap.</b> The 101-column offender contains
+<code>using the heuristic fallback</code>, which is asserted by <b>six</b> test files plus
+<code>tests/bdd/features/explain.feature</code>. Wrapping inserts a newline mid-sentence and
+breaks every one of them; and under pytest <code>term_width()</code> falls back to 80, so the
+wrap fires in the suite rather than only on a narrow terminal. The change is a
+<code>wrap()</code> helper in <code>core/output.py</code> — mutation-gated, so boundary-width
+tests, not coverage — plus re-pinning those assertions. Shortening the strings instead is a
+smaller diff but a voice decision, and the substring pinned by the tests would have to survive
+either way.</p>
+        <div class="meta">
+          <span class="m">Complexity <b>M</b></span>
+          <span class="m">Impact <b>Low</b></span>
+          <span class="m">Wow <b class="wow">★★</b></span>
+          <span class="m">11 commands still emit prose hints past 80 columns; the worst is one string pinned by six test files</span>
+        </div>
+      </article>
       <!-- /ROADMAP:cards -->
 
     </div>

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/tests/functional/test_cli_team.py

@@ -89,6 +89,22 @@ class TestCohort:
         assert data[0]["member"] == _member("pilot", 100)
         assert data[1]["member"] == _member("zero", 0)
 
+    def test_no_literal_double_percent_reaches_the_terminal(self, boost, tapped):
+        """`%%` is printf escaping, and neither of these two strings is ever
+        %-formatted — so both reached the terminal verbatim as `%%`. The
+        membership assertion above stops one character short of the listing's
+        occurrence, which is how it survived.
+        """
+        boost("cohort", "create", "pilot", "--skills", "brainstorming",
+             "--percent", "50")
+        listing = boost("cohort", "list")
+        assert "%%" not in listing.out
+        assert "sha256(user:cohort) % 100 < rollout" in listing.out
+
+        helptext = boost("cohort", "--help")
+        assert "%%" not in helptext.out
+        assert "a 50% rollout" in helptext.out
+
     def test_zero_percent_creates_out_and_apply_skips(self, boost, tapped):
         r = boost("cohort", "create", "zero", "--skills", "brainstorming",
                  "--percent", "0")

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/tests/unit/test_errors_and_cli_table.py

@@ -137,6 +137,42 @@ class TestMainDispatch:
         for _icon, title, _desc in cli.GROUPS.values():
             assert title in out
 
+    @pytest.mark.parametrize("cols", [60, 80, 100, 120])
+    def test_help_never_overflows_the_terminal(self, sandbox, capsys,
+                                               monkeypatch, cols):
+        """The top-level help was the one screen in the CLI that ignored the
+        terminal: `search` drops columns and truncates at 60, while `boost
+        --help` emitted a 105-column banner at every width, wrapping mid-word
+        for anyone not running a wide pane.
+        """
+        monkeypatch.setenv("NO_COLOR", "1")
+        monkeypatch.setattr(cli.out, "term_width", lambda: cols)
+        assert cli.main(["--help"]) == 0
+        for line in capsys.readouterr().out.splitlines():
+            assert cli.out.visible_len(line) <= cols, line
+
+    def test_help_keeps_the_product_name_and_version_at_60_columns(
+            self, sandbox, capsys, monkeypatch):
+        """Fitting the banner must not cost the two facts it carries."""
+        monkeypatch.setenv("NO_COLOR", "1")
+        monkeypatch.setattr(cli.out, "term_width", lambda: 60)
+        assert cli.main(["--help"]) == 0
+        out = capsys.readouterr().out
+        assert "Homebrew for AI coding skills" in out
+        assert cli.__version__ in out
+        assert "80 commands · 8 groups" in out
+
+    def test_help_still_names_every_command_when_narrow(self, sandbox, capsys,
+                                                        monkeypatch):
+        """Summaries may be clipped to fit; command names never are, or the
+        help stops being usable as an index."""
+        monkeypatch.setenv("NO_COLOR", "1")
+        monkeypatch.setattr(cli.out, "term_width", lambda: 60)
+        assert cli.main(["--help"]) == 0
+        out = capsys.readouterr().out
+        for name, _g, _m, _s in cli.COMMANDS:
+            assert "\n  %s" % name in out, name
+
     def test_help_word(self, sandbox, capsys):
         assert cli.main(["help"]) == 0
         assert "80 commands · 8 groups" in capsys.readouterr().out

{boost_skill_cli-1.2.19 → boost_skill_cli-1.2.20}/tests/unit/test_output.py

@@ -457,6 +457,56 @@ class TestPanel:
         assert "\033[1minventory\033[0m" in output.panel("x", title="inventory")
 
 
+class TestPanelFitsTerminal:
+    """`panel()` sized itself to its content and ignored the terminal, so a
+    long line produced a box wider than the pane — and a broken box is the
+    ugliest overflow there is, because the border wraps into the next row.
+    Measured: `boost count` drew 108 columns into an 80-column terminal.
+    """
+
+    def _rows(self, monkeypatch, cols, lines, **kw):
+        monkeypatch.setenv("NO_COLOR", "1")
+        monkeypatch.setattr(output, "term_width", lambda: cols)
+        return output.panel(lines, **kw).split("\n")
+
+    def test_untouched_when_it_already_fits(self, monkeypatch):
+        rows = self._rows(monkeypatch, 40, "x" * 36)
+        assert [output.visible_len(r) for r in rows] == [40, 40, 40]
+        assert "…" not in rows[1]
+
+    def test_exactly_at_the_limit_is_not_clipped(self, monkeypatch):
+        # inner 36 + "│ " + " │" == 40 == the full width: the boundary case a
+        # mutant flipping <= to < would break.
+        rows = self._rows(monkeypatch, 40, "y" * 36)
+        assert rows[1] == "│ " + "y" * 36 + " │"
+
+    def test_one_column_over_is_clipped(self, monkeypatch):
+        rows = self._rows(monkeypatch, 40, "z" * 37)
+        assert [output.visible_len(r) for r in rows] == [40, 40, 40]
+        assert rows[1].endswith("… │")
+
+    def test_every_row_is_the_same_width_after_clipping(self, monkeypatch):
+        rows = self._rows(monkeypatch, 30, ["short", "w" * 90, "mid" * 4])
+        assert len({output.visible_len(r) for r in rows}) == 1
+        assert output.visible_len(rows[0]) <= 30
+
+    def test_title_is_clipped_too(self, monkeypatch):
+        rows = self._rows(monkeypatch, 24, "a" * 50, title="t" * 40)
+        assert all(output.visible_len(r) <= 24 for r in rows)
+        assert len({output.visible_len(r) for r in rows}) == 1
+
+    def test_a_narrow_pane_still_yields_a_box(self, monkeypatch):
+        rows = self._rows(monkeypatch, 8, "content that is far too long")
+        assert len(rows) == 3
+        assert rows[0].startswith("╭") and rows[-1].endswith("╯")
+        assert all(output.visible_len(r) <= 8 for r in rows)
+
+    def test_a_zero_width_pane_does_not_crash_or_go_negative(self, monkeypatch):
+        rows = self._rows(monkeypatch, 0, "anything")
+        assert len(rows) == 3
+        assert all(output.visible_len(r) >= 0 for r in rows)
+
+
 class TestEmptyState:
     def test_message_only_plain(self, monkeypatch):
         monkeypatch.setenv("NO_COLOR", "1")