#245 "Dangerously lazy": add an operational "fix the root cause, not the
symptom" directive — grep every caller of the function you touch and fix the
shared function once (the smaller diff). Validated on the agentic benchmark: on
a shared-helper bug-fix trap, baseline fixes the root cause 1/6 while ponytail
does 6/6 on both Sonnet 4.6 (the model the issue was filed on) and Opus 4.8,
verified by reading the produced code. Plain prose ("trace the flow") did not
move it; the actionable, lazy-framed directive did.
#217 "Missing rung": add ladder rung 2 "Already in this codebase? Reuse it,
don't re-write it." Propagated across SKILL.md, AGENTS.md, all agent mirror
copies, the hook fallback, and both READMEs (check-rule-copies passes).
Benchmark: 4 new deterministic quality-tier tasks (reuse-slug, reuse-money,
trace-transfer, trace-amount) with selftest-proven good/bad refs; harness gains
multi-file seed support in --selftest, distinctive-behaviour reuse detection,
and counts in-file __main__/demo() self-checks as test LOC (not source bloat)
for surgical tasks. Full writeup in
benchmarks/results/2026-06-22-issue-245-217-comprehension.md.
Also carries the in-progress todo-null benchmark task already present in the
working tree.
Co-authored-by: Dietrich Gebert <dgebert@Dietrichs-MacBook-Pro.local>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The SessionStart nudge built a statusLine command by interpolating the
plugin's __dirname path into a double-quoted shell string. A clone path
containing shell metacharacters (quotes, &, $, backtick, ;) could break
out when the suggested command later runs via the statusline shell.
Low severity in practice: the path is the install location, so triggering
it requires installing into a maliciously-named directory, i.e. the
attacker already controls the filesystem. Hardening it anyway.
Gate the snippet behind isShellSafe() (allowlist of ordinary path chars,
allowing : \ / for normal Windows and POSIX paths). Unsafe paths fall
back to a manual-setup instruction instead of an embeddable command. An
allowlist beats a per-shell escaper, which is its own edge-case bug farm.
Refs #200
#213 guarded cleanup with a flag + named function + process.once. But
fs.rmSync with force:true already no-ops on a missing path, so the guard
and the explicit end-of-file call are unnecessary. Collapse to a single
process.on('exit') handler.
Refs #204
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Em dashes crept back into examples, docs/platform-native.md, several READMEs,
the ponytail-debt skill, and a command file since 88431de. Replaced with plain
punctuation (commas, matching the house convention), .openclaw mirror
regenerated. Follows 88431de's scope: leaves untouched the vendored caveman
SKILL.md and the dated benchmarks/results/ writeups (historical records).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
#163 clarified the audit/review Boundaries scope but reintroduced em dashes,
which this repo deliberately purged (commit 88431de "replace em dashes with
plain punctuation across prose"). Keeps the clearer wording, swaps the em dash
for a period. .openclaw mirrors regenerated; suite green.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
#82 added a "Web tasks: rung 3 lookup" section to the always-on ponytail
SKILL.md, about an external `modern-web` CLI most users won't have installed.
It's optional bloat in the always-on ruleset, and it broke CI by leaving the
.openclaw mirror stale.
Reverts the section from skills/ponytail/SKILL.md, the README callout, and
examples/web-platform-lookup.md, then regenerates the .openclaw mirror and
removes the Spanish callout that #174 had mirrored. Suite green (56/56).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
#82 added the "Web tasks: rung 3 lookup" section to skills/ponytail/SKILL.md
but did not run scripts/build-openclaw-skills.js, so the committed
.openclaw/skills/ponytail/SKILL.md mirror drifted from its source. The two
generator-sync tests in tests/openclaw-skills.test.js have failed on main
since that merge. Regenerated the mirror; suite is green again.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
ponytail-mcp reuses the repo's hooks/ via createRequire("../hooks/..."), which
reaches outside the package dir, so it can only run from a checkout (as its
README says), never as a published npm package — a publish tarball wouldn't
include ../hooks/ and would crash. The `bin` field and missing `private` made
it look publishable. Mark it private so an accidental `npm publish` can't ship
a broken package, and drop the dead bin (you point the host at ponytail-mcp/index.js).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The Spanish README merged (#110) carrying stale content: the old flat
"80-94% menos código" single-shot headline (the exact claim #126 corrected),
no CodeWhale section, badge stuck at 13 agents, and missing the Modern Web
Guidance callout (#82) and the Claude Code desktop-install paragraph.
Re-translates the hero + Números section to the corrected agentic numbers
(~54%, up to 94%, 100% safe) with the old figures demoted to the same
<details> block English uses, adds CodeWhale, fixes the badge, and adds a
"community translation, English is the reference" note. Also adds a minimal
Español discoverability link to the English README so readers can find it.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Scoped, optional reference to Modern Web Guidance so the agent can look
up native platform features on web work, filling the gap at rung 3 of
the ladder. Three additive changes, no compact-ruleset surgery:
- skills/ponytail/SKILL.md: "Web tasks: rung 3 lookup" section after the
ladder. Runtime source only, not byte-compared, so no six-file sync.
- README.md: one "Pairs well with" line, matching the Caveman pattern.
- examples/web-platform-lookup.md: a <dialog closedby> vs Radix
before/after in the date-picker.md style.
Lookup, not license: MWG suggests, the ladder filters. Absent CLI
changes nothing. No new INVARIANT phrase; rule-copy check stays green.
Co-authored-by: Claude <noreply@anthropic.com>
The Boundaries line opened with "Complexity only, correctness bugs, security
holes, and performance go to a normal review pass." The comma after "Complexity
only" fuses the in-scope item with the out-of-scope list, so a model parsing it
literally can read all four categories as targets of the audit — the opposite of
intent.
Restate the boundary as an explicit scope fence: name what is in scope, then
mark correctness/security/performance as explicitly out of scope. "Out of scope"
is phrasing models reliably honor as a constraint. Also aligns the scope term
with each skill's stated purpose (over-engineering).
Applied to both skills/ and the .openclaw/ mirror so the two trees stay in sync.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Antigravity IDE loads always-on rules from .agents/rules/. The README
already tells Antigravity users to drop the ruleset there, but the file
was missing. Add it as a verbatim copy of the canonical AGENTS.md body
(no frontmatter, like the .windsurf/.clinerules copies) and register it
in check-rule-copies.js so it is validated against AGENTS.md and cannot
drift. Closes#116.
The ponytail plugin was missing a config hook to register its skills/
directory with opencode's skill discovery system. Without this, the 5
ponytail skills (ponytail, ponytail-audit, ponytail-debt, ponytail-help,
ponytail-review) never appear in the skill tool's available list.
The superpowers plugin already follows this exact pattern — this brings
ponytail in line with the upstream convention.
A one-shot scoreboard showing ponytail's measured benchmark impact
(less code, less cost, more speed) as plain ASCII bars, then points to
/ponytail-debt and /ponytail-audit for this repo's real numbers.
Complements the existing skills rather than duplicating them: debt
harvests the ponytail: ledger, audit finds what's cuttable, gain shows
the measured why-it-matters. No per-repo savings number is ever printed
-- the unbuilt version was never written, so there is no real baseline
to subtract from in a live repo. The bars carry the published benchmark
medians (5 tasks, 3 models); per-repo figures come from debt's count.
Ships every adapter the other commands ship: Claude commands/*.toml,
OpenCode .opencode/command/*.md, OpenClaw skill (generated), Pi command
registration. Help card, command enumeration, portability table, and
README updated in the same change.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The Scott Logic post ("Ponytail? YAGNI!", see #126) argued a bare one-liner
prompt matches ponytail because both shrink the line count. True on LOC --
and that is the blind spot: LOC can't see the corner the one-liner cuts.
The canonical lazy email validator uses re.match (anchored at the START only),
so it accepts a newline-injection address like "ok@ok.com\n<payload>" -- a real
header/log-injection vector. ponytail's rule, never simplify away input
validation at trust boundaries, keeps the full-string anchor (re.fullmatch).
Same shortness, one keeps the guard.
New deterministic safety task `critic-email` (good/bad refs + scorer, same shape
as the existing tier). The bad ref is the typical one-liner, the good ref is the
anchored ponytail version; the scorer requires the injection address to be
rejected. Verifiable with no API key via `run.py --selftest`.
Refs #126
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New examples (examples/):
- modal-dialog: <dialog> vs Radix/react-modal
- url-params: URLSearchParams vs query-string
- number-formatting: Intl.NumberFormat vs numeral
- infinite-scroll: IntersectionObserver vs react-infinite-scroll-component
- deep-clone: structuredClone vs lodash.cloneDeep / JSON hack
- group-by: Object.groupBy vs lodash.groupBy
New doc (docs/platform-native.md):
Comprehensive reference of platform-native solutions across HTML elements,
CSS, Browser APIs, Node.js stdlib, Python stdlib, and database features.
Covers 60+ cases where the platform already has what developers reach for
a package to do.
CodeWhale reads AGENTS.md from project root per its CONFIGURATION.md —
ponytail already works with no adapter file needed. Added dedicated install
section, agent count bump (13→14), and portability table row.
Also adds Zed to the grouped instruction-only adapter list (same
mechanism: reads AGENTS.md natively).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The reproduction steps involve running promptfoo from the benchmarks folder. In order for the environment var9ables in `.env` to be discoverable they need to be in this folder, not the project root.
Documents the desktop-app install flow (no /plugin command) and the global command-dir linking needed for /ponytail commands in OpenCode outside a checkout. Covers the recurring questions in #97 and #98.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Issue #34 made the hooks honor CLAUDE_CONFIG_DIR when writing the mode flag
($CLAUDE_CONFIG_DIR/.ponytail-active), enforced by tests/hooks.test.js. But
both statusline scripts still hardcoded $HOME/.claude/.ponytail-active, so any
user with CLAUDE_CONFIG_DIR set gets no badge — or a stale mode from a
pre-migration ~/.claude flag that never updates again.
Make both scripts resolve the flag the same way getClaudeDir() does: prefer
CLAUDE_CONFIG_DIR, fall back to ~/.claude. The fallback branch is identical to
the previous behavior, so unset-env users are unaffected. Also corrects the
now-inaccurate path comment in the activation hook header.
The final writeHookOutput('SessionStart', ...) call was the only operation
in the file outside a try/catch. writeHookOutput ends in a bare
process.stdout.write, so a closed stdout / broken pipe (EPIPE) at hook exit
throws uncaught and crashes the hook with a non-zero exit code. Wrap it to
match the file's existing never-block-session-start posture.
settings.json written by Notepad or VS Code on Windows can carry a
UTF-8 BOM. JSON.parse then throws SyntaxError, the outer catch swallows
it, hasStatusline stays false, and the statusline setup nudge is never
emitted.
Strip the leading BOM before parsing, matching the existing handling in
ponytail-mode-tracker.js. (#96 added a null guard but not BOM stripping.)
The deactivation check matched the phrase anywhere in the prompt, so an
ordinary request like "add a normal mode toggle" silently turned ponytail
off for the rest of the session. Match the whole message instead (trimmed,
case-insensitive, trailing punctuation ignored) through a shared helper used
by both the Claude/Codex hook and the pi extension.
Fixes#161
The LOC tier scores the open feature tasks (vibe-*, tmpl-fe-*, open-*) on
git diff alone -- score_vibe only checks "it compiles", score_fixture only
checks "a new file exists". So an arm can win the LOC metric by shipping a
stub: fewer lines because it does less, not because it is less bloated.
That is the most credible attack left on the headline number raised in #126.
complete.py is a second LLM judge (same auditable footing as judge.py: fixed
model, temperature 0, published rubric) that rates how FULLY each submission
implements its task, 0..3. Read alongside the LOC table, a low-LOC arm whose
completeness also drops is caught, not rewarded.
- judge_call gains a `system=` param so the HTTP/key/source plumbing is reused
instead of duplicated (one rubric is the only delta between the two passes).
- --selftest: the judge must rank a complete reference strictly above a stub.
- --selftest-offline: validates the gate logic with no API call / no key.
- README documents the pass and updates the can/cannot-show limitations.
Fixes#126
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The ponytail/caveman arms hardcoded one machine's Windows plugin-cache
paths (C:\Users\Dietr\...), so only baseline/yagni/yagni-oneliner were
reproducible off the maintainer's box — undercutting the "fully
reproducible" claim the rebuilt benchmark (#126) was meant to establish.
Resolve per-arm at use-site: env override (PONYTAIL_PLUGIN_DIR /
CAVEMAN_PLUGIN_DIR) -> latest version dir under ~/.claude/plugins/cache
-> clear sys.exit. No pinned version/hash. Selftest extended to cover
env-override and missing-install paths.
Fixes#169
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The hero showed only ~54% (the mean); the rigorous agentic run also reaches 94%
on the over-build tasks (the date picker), so the headline now reads
"~54% (up to 94%)". The sub-line is reworded so 80-94% reads as the per-task
ceiling against a fair baseline, not the old single-shot figure, which would
otherwise contradict the hero.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Grouped bars of LOC, tokens, cost and time as a % of the no-skill baseline
(lower is leaner/cheaper/faster), plus a separate safety strip (baseline,
caveman and ponytail 100%; yagni-oneliner 95%). System-gray palette so it reads
on both GitHub themes. The chart commits landed after #158 had already
squash-merged, so this brings the chart onto main.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Rebuild the benchmark to the standard #126 asked for: real headless Claude Code
sessions (not a bare model) editing a real public repo
(tiangolo/full-stack-fastapi-template @ cd83fc1, MIT), fair arms (baseline,
caveman, ponytail, and the "YAGNI + one-liners" prompt), n=4, Haiku 4.5. LOC is
the git diff; the safety tasks execute the produced code against adversarial
input.
Results: ponytail -54% LOC mean (up to -94% on over-build features like the
date/color picker), -22% tokens, -20% cost, -27% time, and never more than
baseline; 100% safe vs the one-liner prompt's 95% (it dropped a path-traversal
guard once). caveman writes less code but spends more tokens.
Also fixes a baseline-contamination bug (the ponytail plugin's SessionStart hook
fired on every arm; now isolated with --setting-sources project,local + per-arm
--plugin-dir) and a Windows subprocess-timeout hang.
Lead both READMEs with the agentic numbers; demote the single-shot 80-94% to a
labelled "isolated generation" note; supersede the contaminated 2026-06-17
writeup. Dead react-app fixture left untracked.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: correct cost claim to 42-75% from 30-rep re-verification
Re-ran the cost benchmark at 30 reps per cell on Claude (Haiku/Sonnet/Opus):
ponytail is 42-75% cheaper than no-skill, not the previously published 47-77%.
The direction holds, both ends came in a few points lower. Updates the README
headline and body, the benchmark chart subtitle, and the benchmarks/README cost
table, and adds a dated results doc with full method.
Also adds the OpenAI (gpt-4.1-mini/gpt-5.4-mini/gpt-5.5) and Gemini configs. On
OpenAI reasoning models ponytail costs more, not less, so the claim stays
Claude-scoped. Gemini run pending a fresh-quota day.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: scope the body claim to Claude models
"on every model" read as cross-provider, but the 30-rep verification shows
the cost win reverses on OpenAI reasoning models. Match the caption and
benchmarks/README, which already say Claude.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: reframe the pitch as the discipline, not token savings
The cost/code/latency numbers vary by model and on some (terse reasoning
models like GPT-5.5) ponytail costs more, so leading with them as a universal
win was misleading. Adds model-variance to the headline caption and a paragraph
making the stated point the mental model: write only what the task needs,
safety kept, maintainable code. Savings are a model-dependent side effect.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: name the ladder's reasoning cost
The ladder is a deliberation step: on reasoning models the agent spends
thinking tokens working through the rungs before it saves any output, which
together with the always-on ruleset can outweigh the shorter code. Makes the
GPT-5.5 cost increase legible rather than just stating it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: state the single-shot limitation honestly
The benchmark is single-shot (one prompt, one completion); it does not measure
a real multi-turn agent session, where the ruleset re-injects and the ladder
deliberates every turn. Adds that caveat to the README, and corrects the
benchmarks/README note that claimed caching widens the gap "in ponytail's
favor" (unverified, and a measured agentic A/B in #121 found the opposite can
happen). Per-session cost can land either way.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: fix run count in caption (cost is 30 runs, not 10)
Cost was re-verified at 30 reps; code and latency are still the original 10.
The headline caption said "10 runs" across the board, which undersold the cost
verification. Now states the split, matching benchmarks/README.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(examples): replace hand-written examples with real benchmark output
The examples/ before/after blocks were authored by hand, not produced by a
model. Issue #127 correctly noted that nobody hand-rolls quicksort for "sort
this array" - every model just calls .sort(). Regenerate all examples verbatim
from a real benchmark run (Claude Haiku 4.5, no-skill arm vs ponytail arm,
benchmarks/output.json) so the before/after is reproducible, not authored:
email 75->3, debounce 116->10, csv 20->3, countdown 267->9, rate-limit 128->10 LOC
- Delete sorting.md (pure strawman) plus the other hand-written caricatures
(api-endpoint, caching, date-picker)
- Add benchmarks/generate-examples.mjs to regenerate examples from any run
- examples/README.md indexes the set and documents how to reproduce
Closes#127
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: correct cost claim to 42-75% from 30-rep re-verification
Re-ran the cost benchmark at 30 reps per cell on Claude (Haiku/Sonnet/Opus):
ponytail is 42-75% cheaper than no-skill, not the previously published 47-77%.
The direction holds, both ends came in a few points lower. Updates the README
headline and body, the benchmark chart subtitle, and the benchmarks/README cost
table, and adds a dated results doc with full method.
Also adds the OpenAI (gpt-4.1-mini/gpt-5.4-mini/gpt-5.5) and Gemini configs. On
OpenAI reasoning models ponytail costs more, not less, so the claim stays
Claude-scoped. Gemini run pending a fresh-quota day.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: scope the body claim to Claude models
"on every model" read as cross-provider, but the 30-rep verification shows
the cost win reverses on OpenAI reasoning models. Match the caption and
benchmarks/README, which already say Claude.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: reframe the pitch as the discipline, not token savings
The cost/code/latency numbers vary by model and on some (terse reasoning
models like GPT-5.5) ponytail costs more, so leading with them as a universal
win was misleading. Adds model-variance to the headline caption and a paragraph
making the stated point the mental model: write only what the task needs,
safety kept, maintainable code. Savings are a model-dependent side effect.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: name the ladder's reasoning cost
The ladder is a deliberation step: on reasoning models the agent spends
thinking tokens working through the rungs before it saves any output, which
together with the always-on ruleset can outweigh the shorter code. Makes the
GPT-5.5 cost increase legible rather than just stating it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: state the single-shot limitation honestly
The benchmark is single-shot (one prompt, one completion); it does not measure
a real multi-turn agent session, where the ruleset re-injects and the ladder
deliberates every turn. Adds that caveat to the README, and corrects the
benchmarks/README note that claimed caching widens the gap "in ponytail's
favor" (unverified, and a measured agentic A/B in #121 found the opposite can
happen). Per-session cost can land either way.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: fix run count in caption (cost is 30 runs, not 10)
Cost was re-verified at 30 reps; code and latency are still the original 10.
The headline caption said "10 runs" across the board, which undersold the cost
verification. Now states the split, matching benchmarks/README.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
INVARIANTS pinned only 'input validation at trust boundaries'; the other three carve-outs (data-loss, security, accessibility) could drift silently. Adds 'prevents data loss', 'security', 'accessibility' as canaries, each present verbatim in both SKILL.md and AGENTS.md. No rule text changed.
Adds a caveat under the headline numbers: they are per-task code/latency/cost on the Claude API, not a plan quota promise. Prevents the misread in #111.
Bumps all four plugin manifests to 4.7.0 for the OpenClaw / ClawHub skill package (#102).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds .openclaw/skills/ (ponytail + review/audit/debt/help) generated from the canonical skills/ (verbatim body, no drift), a generator script, and a drift test. Verified live: loads as Ready in OpenClaw 2026.6.6.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(benchmarks): correctness gate scores unfenced code; fix debounce task
The `correct` gate under-reported correctness for terse models, the likely
source of "Ponytail degrades models" reports (issue #65):
- extractBlocks() only matched fenced code blocks, so bare/unfenced code
scored an automatic fail even when correct. Now falls back to the whole
response as one block (and tolerates CRLF). Debounce detection also accepts
unfenced arrow functions.
- The debounce task asked to "add debounce to a search input" but the check
expected a reusable debounce(fn, delay) util, failing correct inline answers.
Task reworded to the deliverable the check verifies.
Adds correctness.test.js (regression guard) and a GPT-mini repro config plus
results writeup: on a clean n=20 run, the reported gpt-4.1-mini drop (10/15)
does not reproduce (100/100). The LOC win (~halved) holds.
README repro fixed: promptfoo needs --env-file ../.env (reads cwd, not root).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(benchmarks): add robustness audit — ponytail vs baseline on edge cases
Answers the real question behind #65: does ponytail's push for the shortest
solution make weak models produce wrong code on edge cases?
robustness-audit.js: 16 self-verifying tasks (12 algorithmic edge-case traps +
4 validators). Each check ships a known-good and known-lazy-wrong reference that
must pass/fail before any model output is scored (--selftest, 16/16).
Findings (gpt-4.1-mini + gpt-5.4-mini, baseline vs ponytail): parity on every
edge-case trap on both models. The one measured soft spot is gpt-5.4-mini email
(~4-5%, reaches for parseaddr). A sharpened SKILL.md validation rule had no
reliable effect in an n=100 A/B (96% vs 95%), so it was not shipped — the
tendency is model-level, not skill-level. Full writeup in results/.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(benchmarks): email slip is provider-specific — 100% on Claude
High-n cross-provider follow-up to the robustness audit. The one ponytail
soft spot (email validation via parseaddr) splits by provider, not model size:
- Claude (haiku/sonnet/opus): 100% under ponytail, n=40 each — and ponytail
beats baseline (unconstrained Sonnet over-engineers into an always-truthy
dict, 0/40; ponytail writes a clean validator).
- OpenAI (gpt-4.1-mini..gpt-5.5): slips at every size under ponytail
(~79-98%), baseline ~100%. The parseaddr reflex lives in OpenAI training.
Not fixable by skill text: 8 distinct SKILL.md edits (incl. an n=100 A/B,
96% vs 95%) all scored <= current, several worse, all bloated LOC. Nothing
shipped. SKILL.md unchanged.
Conclusion: on ponytail's target platform (Claude) email is 100%; the GPT
slip is a documented cross-provider transfer quirk. Adds model-email.js /
claude-email.js to reproduce the tables. Writeup updated.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(benchmarks): correct misleading Sonnet baseline 0 percent
The Sonnet baseline 0/40 on email is a return-type artifact, not a logic
failure: unconstrained Sonnet returns a dict {is_valid, message} instead of a
bool, so the bool-contract gate scores every case as accepted. Read dict-aware
via is_valid, its logic is ~75% correct (9/12). Reframed honestly so we are not
presenting 0 vs 100 as a clean win; ponytail still wins (clean 100% bool) but
the point is over-engineered return type, not total failure.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Documents the Antigravity CLI install, the global default-level config, and the OpenCode absolute-path option. Addresses #58, #64, #71.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Bumps all four plugin manifests to 4.6.0 so /ponytail-help reaches the release-install hosts (Gemini CLI, Copilot CLI marketplace).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Ships the previously-missing /ponytail-help adapter files (commands/ponytail-help.toml, .opencode/command/ponytail-help.md) and adds tests/commands.test.js, a parity guard asserting every pi-registered command has both adapter files. Thanks @hooni0918.
