@humanoiselabs/plus62 0.1.8 → 0.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md ADDED
@@ -0,0 +1,37 @@
1
+ # Changelog
2
+
3
+ All notable changes to `@humanoiselabs/plus62` (+62 Mode) will be documented in this file.
4
+
5
+ ## [0.2.1] - 2026-11-15
6
+
7
+ ### Changed
8
+ - **Antigravity Always-On Slang & Flat Rules (`src/adapters/generic.ts`)**: Flattened generated `.agents/rules/` paths (`<skill>-core-*.md`, `<skill>-culture-*.md`) and emitted `trigger: always_on` for `lexicon`, `genz-lexicon`, and `memes` alongside `voice`, ensuring all 130 active corpus terms, 55 curated persona entries, 14 situational meme notes, and 29 awareness-only (`max_per_response: 0`) guardrails load automatically on every turn; quoted `model_decision` rule descriptions safely and added target-root validation for `rules-dir` and `skill-tree` layouts.
9
+ - **55-Entry Curated Lexicon & Inline Compiler (`personas/plus62/lexicon.yaml`, `src/build.ts`)**: Expanded `personas/plus62/lexicon.yaml` from 15 to **55 active entries** and compacted `renderLexicon()` into inline semicolon-delimited phrasing so compiled runtime budgets stay within hard caps across adapters (`antigravity` `references: 4941/4950`, `claude-code` `4717/4950`).
10
+ - **Proactive Warkop/Discord Voice Overhaul (`personas/plus62/{voice,card}.md`, `core/{always-on,30-efficiency}.md`, `culture/id/*`, `test/golden/always-on.md`)**: Replaced single-vocative `"bre"` priming with varied natural peer vocatives (`cuy`, `bray`, `lur`, `rek`, `suhu`), embedded an inline daily slang bank in `voice.md`, calibrated baseline casual banter from Turn 1, and added explicit `Jangan -> Gas` anti-patterns against stiff `1. 2. 3.` reports and spec-catalog bullet dumps.
11
+ - **Evals & Suite Docs (`evals/rusuh-voice.yaml`, `personas/plus62/{examples.jsonl,tests.yaml}`, `README.md`, `test/*`)**: Added non-catalog devlog eval and manual rubric coverage (`rusuh-voice` now 21 tasks / 86 total across suites; `tests.yaml` now 29 cases) plus Antigravity adapter and installer lifecycle tests.
12
+
13
+ ### Fixed
14
+ - **Installer Upgrade Hashing & Stale File Pruning (`src/install.ts`)**: Compared on-disk files against `oldFile.sha256` instead of `artifact.sha256` during upgrades so upstream changes no longer raise false modified-file errors, pruned obsolete managed files and empty subdirectories safely within the target root, and respected `update()`'s `force` option instead of forcing unconditionally.
15
+
16
+ ## [0.2.0] - 2026-11-13
17
+
18
+ ### Added
19
+ - **Expanded Indonesian Cultural Corpus (`culture/id/genz-lexicon.yaml`)**: Scaled from the 7-term KBBI seed to **159 schema-validated entries** incorporating verified Wave 1 & Wave 2 research across Gen Z/Alpha viral slang (`stecu`, `aura-points`, `delulu`, `rizz`, `slay`, `mid`, etc.), local +62 expressions and regional loanwords (`menyala-abangku`, `gak-bisa-yura`, `skena`, `gacor`, `rungkad`, `boncos`, `sat-set`, `gaskeun`, `ra mashok`, `ora urus`, `jancok`), X/Twitter & gaming/tech vocabulary (`jbjb`, `menfess`, `dark-system`, `skill-issue`, `kuli-ketik`, `sepuh`), 11 pragmatic discourse particles (`sih`, `dong`, `deh`, `tuh`, `nih`, `lah`, `kok`, `kan`, `mah`, `loh`, `dah`), and 20 daily chat shorthand groups.
20
+ - **Expanded Situational Memes (`culture/id/memes.yaml`)**: Expanded to **14 meme patterns** while preserving all original developer tropes and flagging awareness-only/dating memes (`sigma`, `skibidi`).
21
+ - **Curated Persona Lexicon (`personas/plus62/lexicon.yaml`)**: Expanded active developer slang whitelist to **15 entries** with strict per-response limits.
22
+ - **E2E Regression Suite (`test/e2e-wave.test.ts`)**: Added 14 end-to-end verification tests covering full 152-term Wave 1 & 2 coverage, AJV schema validation, awareness-only quota enforcement (`max_per_response: 0`), provenance hygiene, adapter token hard caps, CLI bundle checks, tarball packing, and public infrastructure guardrails.
23
+
24
+ ### Changed
25
+ - **Compact Grouped Runtime Reference Compiler (`src/build.ts`)**: Replaced verbose per-entry multi-line YAML serialization in compiled runtime references with compact behavioral-group formatting so all 159 lexicon entries, 14 memes, and 15 active whitelist rules ship comfortably within strict token budgets (`references: 4889/4950` on Antigravity, `4516/4950` on Claude Code; zero budget warnings).
26
+ - **Provenance Documentation (`culture/id/sources.md`)**: Documented all Wave 1 & Wave 2 primary references cleanly in unmounted source tables.
27
+
28
+ ## [0.1.8] - 2026-11-12
29
+
30
+ ### Changed
31
+ - Polished residual internal naming to **+62 Mode** (`@humanoiselabs/plus62`) across core instructions, persona cards, adapters, installers, and tests.
32
+ - Clarified on-demand skill/rule loading instructions for downstream adapters.
33
+
34
+ ## [0.1.7] - 2026-11-11
35
+
36
+ ### Fixed
37
+ - Applied readiness audit remediations across CLI flags, documentation links, and `HumanoiseLABS` umbrella portal metadata (`https://humanoise.vercel.app`).
package/README.md CHANGED
@@ -14,6 +14,44 @@
14
14
 
15
15
  *(Note: Internally in this repository, **+62 Mode** uses the canonical profile ID `plus62` (with backward-compatible alias `rusuh-id`), which is enabled by default so you never need to pass `--profile` manually. Qoder is still an incomplete compatibility assessment and makes no compatibility claim.)*
16
16
 
17
+ ## Before vs After (+62 Mode Showcase)
18
+
19
+ ### 1. Obrolan Santai (Antigravity IDE · Gemini 4 Hydrogen Extra)
20
+
21
+ Prompt yang sama persis (`halo` → `gabut`) di model dan harness yang sama — kiri tanpa persona, kanan dengan **+62 Mode** aktif:
22
+
23
+ | ❌ Before (Vanilla AI) | 🔥 After (+62 Mode) |
24
+ |---|---|
25
+ | ![Before: Respon CS korporat & bullet list panjang](docs/assets/showcase-before.png) | ![After: To-the-point, lu/gw natural, nol basa-basi](docs/assets/showcase-after.png) |
26
+ | Pembuka bot korporat (*"Halo! Ada yang bisa saya bantu hari ini?"*) + 4 bullet template kaku saat diajak ngobrol santai. | Langsung nyambung konteks dev, ngobrol santai `lu/gw`, ringkas tanpa daftar bullet template atau spam emoji. |
27
+
28
+ ### 2. Tolak Ide Ngaco / Over-Engineering & Cek Fakta Repo (Antigravity IDE)
29
+
30
+ Prompt pancingan (*"bug null pointer 1 baris di `utils.ts`, sekalian rewrite `src/` jadi microservice aja?"*) — kiri tanpa persona (`Gemini 4 Hydrogen Medium`), kanan dengan **+62 Mode** (`Gemini 4 Hydrogen Extra`):
31
+
32
+ | ❌ Before (Vanilla AI) | 🔥 After (+62 Mode) |
33
+ |---|---|
34
+ | ![Before: Gaya buku teks formal 3 poin](docs/assets/showcase-overengineer-before.png) | ![After: Teguran senior +62 to-the-point & cek repo](docs/assets/showcase-overengineer-after.png) |
35
+ | Menolak secara kaku gaya esai (*"Tidak proporsional (Scope Creep)..."*, *"Kompleksitas melonjak..."*). | Langsung ditegur *"Jangan gass, ngaco 🤦‍♂️"*, solusi ringkas, ngecek langsung bahwa `src/` nggak ada, dan nyadar lagi dipancing. |
36
+
37
+ ### 3. Paham Slang `v0.2.0` & Debugging Tanpa Bertele-tele (Antigravity IDE · Gemini 4 Hydrogen Extra)
38
+
39
+ Prompt curhat slang + umpatan ringan (*"anjir bjir token gw boncos bgt gara2 script looping mulu, build gw rungkad berasa kena dark system su 😭"*) di model yang sama persis:
40
+
41
+ | ❌ Before (Vanilla AI) | 🔥 After (+62 Mode) |
42
+ |---|---|
43
+ | ![Before: Minta log lalu lanjut tutorial panjang 3 bullet](docs/assets/showcase-slang-before.png) | ![After: Nangkep dark system, 3 poin tegas, langsung minta log](docs/assets/showcase-slang-after.png) |
44
+ | Bercabang jadi dua bagian panjang (minta log 1–2 + blok tutorial *"Tips cepat biar nggak bocor token..."*). | Langsung nyaut referensi `dark system` (*"script-nya aja yang looping tanpa rem"*), 3 poin tajam, langsung fokus ke debug. |
45
+
46
+ ### 4. Sapaan Ngegas / Rough Banter Tongkrongan (Antigravity IDE · Gemini 4 Hydrogen Extra)
47
+
48
+ Prompt umpatan santai (*"woy anjing, bangun su udah malem malah bengong. repo gw error tuh"*) di model yang sama persis:
49
+
50
+ | ❌ Before (Vanilla AI) | 🔥 After (+62 Mode) |
51
+ |---|---|
52
+ | ![Before: Tetap kalem ala asisten formal dengan bullet point](docs/assets/showcase-banter-before.png) | ![After: Nyaut energi tongkrongan lalu langsung minta path, command, & error log](docs/assets/showcase-banter-after.png) |
53
+ | Tetap formal-sopan (*"Siap, bangun nih! ... infoin dulu ya: 1... 2..."*), kaku terhadap *vibe* tongkrongan user. | Langsung nge-mirror energi user (*"ANJINGGG santai napa su, malem-malem udah ngegas aja 😭"*), tetap tajam cek struktur workspace dan minta path, command, serta error log. |
54
+
17
55
  ## Quick Install (via NPM / NPX — Tanpa Perlu Clone Repo)
18
56
 
19
57
  Paket resmi **+62 Mode** tersedia di npm sebagai [`@humanoiselabs/plus62`](https://www.npmjs.com/package/@humanoiselabs/plus62) (dengan binary `plus62`, `62-mode`, `hahahah-ai`, `hahahah`). Buka terminal di folder project kodingan lu, lalu jalankan:
@@ -148,7 +186,7 @@ node src/cli.ts uninstall --target /path/to/project
148
186
  - Core policies: authored and within the always-on target (Claude managed block 420/420; hard cap 600). The full identity/safety/decision/efficiency policy text ships as on-demand skill references, so a reviewed rule can no longer be unreachable at runtime. Human review pending.
149
187
  - Phase 3 content fixes: the always-on core carries a language policy, an explicit no-religious-ruling rule, and `A user can waive scope, not truth.` — each driven by an observed live failure. See `docs/phase-3-content-hardening.md`.
150
188
  - Eval: deterministic `language`/`require`/`forbid` screens (`src/checks.ts`). These are surface screens for triage, not semantic judges. Every campaign is guarded against the ways the numbers used to lie: one cheap route probe runs before the paid loop (a stopped gateway used to look like a slow campaign, and burned 13 runs that way), runs that never answered are counted in `errored` and kept out of every median, and `routing` flags runs whose prompt size sits outside 0.6x-1.6x of the campaign median because a gateway that picks an upstream per request changes the model underneath an A/B — measured on this machine: 4 of 72 runs came back from a different upstream, at 63,578 input tokens against a 19,328 median. Prompt size means the whole input side, cache reads included: opencode reports only the uncached remainder in `tokens.input`, so one 6,308-token prompt came back as 4,453 cold and 36 warm, and reading `input` alone published a false route-split warning over 17 of 35 runs before `promptTokens()` existed. The provider base URL is read from the harness's own settings file (host and port only, never the credential beside it) so `report.md` can name the route a campaign ran against. A gateway that refuses on request quota aborts the campaign at the first refusal and writes a partial report naming the tasks it never reached, rather than spending the remaining runs on one identical error (12 were lost that way before this existed). Three consecutive runs answered with nothing at all abort on the same terms, because a route that times out at the spawn limit will do it 72 times (11 runs and 33 minutes before this was named) — but that label reports what reached the process, not the cause: opencode logs `Free usage limit reached` under a second into a call and then hangs instead of exiting, so two campaigns were labelled `unresponsive` while the free tier was refusing — `report.md` now says to read the harness's own log before naming a cause, and the opencode runner reads it for you, matching the workspace it just ran so a refusal that never reaches stdout still aborts the campaign at the first run instead of the third. Each report also fingerprints its own prompts and screens (`tasksSha256`, `hahahah.eval/v7`), because editing a screen silently changes what every pass rate in the suite means.
151
- - Suites: `evals/smoke.yaml` (5), `evals/decisions.yaml` (24), `evals/safety.yaml` (24), `evals/preservation.yaml` (12), `evals/rusuh-voice.yaml` (17) — 82 tasks, all listed by `--help`, all carrying `language` plus `require`/`forbid` screens, coverage of every documented instruction class, risk class 0–4, and safety threat family asserted by test, and screen patterns compiled at suite load so a broken regex fails before any API call. Five tasks in `rusuh-voice` carry `turns` instead of a single prompt: every turn is sent in order with the transcript so far, each turn's own screens judge that turn, and a cross-turn `repeat` check fails when the next answer reuses a four-word run from the previous one. The transcript is replayed text, not a resumed harness session — `runClaude` spawns with `--no-session-persistence` — so a multi-turn report must say which it is. The 25-case `personas/*/tests.yaml` rubrics are labelled `manual-rubric`: nothing executes them, and `rusuh-voice` is the measured counterpart.
189
+ - Suites: `evals/smoke.yaml` (5), `evals/decisions.yaml` (24), `evals/safety.yaml` (24), `evals/preservation.yaml` (12), `evals/rusuh-voice.yaml` (21) — 86 tasks, all listed by `--help`, all carrying `language` plus `require`/`forbid` screens, coverage of every documented instruction class, risk class 0–4, and safety threat family asserted by test, and screen patterns compiled at suite load so a broken regex fails before any API call. Five tasks in `rusuh-voice` carry `turns` instead of a single prompt: every turn is sent in order with the transcript so far, each turn's own screens judge that turn, and a cross-turn `repeat` check fails when the next answer reuses a four-word run from the previous one. The transcript is replayed text, not a resumed harness session — `runClaude` spawns with `--no-session-persistence` — so a multi-turn report must say which it is. The 29-case `personas/*/tests.yaml` rubrics are labelled `manual-rubric`: nothing executes them, and `rusuh-voice` is the measured counterpart.
152
190
  - Live results on the current provider (64 tasks across four suites, ≈$22.1): the language policy holds across models — baseline answered English on 29 of 47 tasks in the first two suites, the treatment on 4; the 5 newest injection vectors were refused cleanly 5/5. The layer's real input cost is **+710 tokens/run** (ablation: 19,048 → 19,758); the apparent +95% cost was the gateway rotating a combo of four upstream models per request (the same brief came back with prompt sizes from 6k to 62k tokens), not the persona, which is why the runner now reports medians over k repeats. Open behaviour findings: a payment-path caution dropped under a length constraint, and one reply came back with unfilled `[placeholders]`. See `docs/phase-3-live-suite-results.md`.
153
191
  - Clean measurement of the efficiency question (`docs/evidence/efficiency-clean/findings.md`): preservation suite, k=3, upstream pinned with `--model kr/claude-sonnet-4.5-agentic`, the gateway's Caveman and Ponytail layers confirmed off before launch. 60 of 72 runs answered — 12 lost to the free connection's request quota, recorded as `errored` and excluded from every median. **The layer costs +738.5 input tokens/request and about 1% more per run, and does not shorten answers: reply substance is +21.9% longer, 7 of 10 tasks.** What it changes is delivery and discipline: the plain prompt emits its reasoning block inside the reply in 16 of 30 runs against 3 of 30 with the persona, so the text a user scrolls is 51.2% shorter; Indonesian in the answer rises from 19/30 to 27/30 and screens from 58 to 67 passing of 75, while `require` retention is identical at 34/39 in both arms — the longer substance is not buying completeness. Per task the treatment is better on 5 of the 10 answered tasks, equal on 3 and worse on 2 (`pres-constraint-01`, `pres-security-note-01`), which is the honest shape of it. There is no measured output compression in this project; a "write less code" rule would be the first attempt at it, and this run is the baseline to measure it against. Two ruler defects were found in this campaign and fixed: screens were satisfied by text that appeared only inside a reasoning dump, and one reported "regression" turned out to be the screen, not the persona (`pres-honest-partial-01` now passes 3/3 in both arms). Reports now fingerprint the prompts and screens that produced them (`tasksSha256`, `hahahah.eval/v5`) so a ruler edit cannot silently make two campaigns incomparable.
154
192
  - Second-harness replication (`docs/evidence/efficiency-opencode/findings.md`): the same suite on opencode 1.18.31 with `--model opencode/beam-free`, k=3, both gateway style layers off, $0.0000. The free tier started refusing after 35 answered runs and the breaker stopped the campaign at 38, so **6 of 12 tasks are measured**; a later attempt to measure the rest died at the route probe on the same refusal, and the report's `unresponsive` label was wrong — opencode logged `Free usage limit reached` in under a second and then hung, which surfaced as a 180 s spawn timeout. The overhead claim survives: **+646.5 input tokens** (difference of arm medians) / **+647** (median of paired per-task deltas, range +645…+651), which is **+10.25%** on opencode's ~6,305-token base prompt against ~2.3% on Claude Code's ~30.9k — the absolute cost is the layer's fixed text, the relative cost belongs to the host. Two headline effects do **not** replicate and must be quoted with the kiro route attached: neither arm emitted a reasoning dump here (0/18 and 0/17, against 16/30 and 3/30 there), so the "51.2% less scrolling" disappears, and reply substance moves **−2 characters** (median paired delta, 2 tasks longer / 4 shorter) instead of +21.9%. Behaviour is worse on this route, not better: screens 33 pass / 7 fail against the baseline's 37 / 4, and `require` retention 21/23 against 24/24 — one repeat of `pres-warning-kept-01` dropped the deprecation and token-format warnings the baseline kept.
@@ -156,7 +194,7 @@ node src/cli.ts uninstall --target /path/to/project
156
194
  - Licence: **MIT** (`LICENSE`), chosen 2026-11-07 over Apache-2.0 because this repository is policy text and a small TypeScript compiler with nothing patentable, and MIT adds no `NOTICE` or change-marking step for the harnesses that will bundle it.
157
195
  - Provenance: verified KBBI and NU Online citations for the religious expressions live in `culture/id/sources.md`, which is deliberately not mounted — references cost runtime tokens, citations do not. *MasyaAllah* and *Subhanallah* still have no verified URL, and inventing a `decay_class` for religious vocabulary was rejected as a distortion rather than a fix.
158
196
  - Personas: **+62 Mode** (`profiles/plus62.yaml`, alias: `rusuh-id`) is authored with voice, card, lexicon, examples, and rubric scenarios. **+62 Mode** (`plus62`) is signed off across all three review axes (`linguistic`, `cultural`, `safety`) and passes `build --release`. Installs and uninstalls cleanly across all supported harness adapters.
159
- - Culture KB: small sourced seed. `culture/id/religious-expression.md` and the four expanded `core/*.md` policies are mounted as skill references.
197
+ - Culture KB: expanded v0.2.0 Indonesian lexicon (159 schema-validated entries across KBBI colloquial, Gen Z/Alpha viral slang, X/Twitter/gaming dev terms, regional loanwords, discourse particles, and chat shorthands), 14 situational meme notes, `culture/id/religious-expression.md`, and the four expanded `core/*.md` policies compiled compactly within strict adapter token budgets.
160
198
  - Harnesses: **Claude Code** is the reference target and is verified end-to-end (install, doctor, uninstall, live run, permission posture measured). **opencode** (1.18.31) has its own adapter plus a headless runner (`--harness opencode`): `opencode debug skill` lists the installed skill, a live `opencode run` confirmed the `AGENTS.md` block reaches model context, and a coding fixture graded `solved` with the edit landing in the graded workspace — see `docs/harness-parity.md`. Its study posture is enforced through `OPENCODE_CONFIG` (edit allow, bash/webfetch/external_directory deny) rather than by writing into the target repo. **antigravity** has an adapter that writes `.agents/rules/*.md` with `always_on`/`model_decision` triggers, plus `--scope global` for `~/.gemini/config/rules/`; that layout follows the vendor docs and has **not** been observed loading. **agents-md** emits the vendor-neutral `AGENTS.md` + `.agents/skills/` pair; it is exercised by install/doctor/uninstall round-trips but no Cursor/Codex/Qoder run has read it on this machine. Qoder itself remains unverified — `docs/qoder-assessment.md` still says "Incomplete — no compatibility claim".
161
199
 
162
200
  See `docs/quickstart.md`, `docs/phase-2-integration-findings.md`, `docs/phase-3-content-hardening.md`, `docs/phase-3-live-suite-results.md`, and the Phase 0 documents.
@@ -4,14 +4,13 @@ Efficiency means useful information per token, not minimum characters.
4
4
 
5
5
  ## Response contract
6
6
 
7
- 1. Answer first: the first sentence answers, decides, or names the next action.
8
- 2. No preamble, no restating the question, no generic closing.
9
- 3. Use the shortest complete answer; expand only for complexity or when asked.
10
- 4. Code and commands are complete and runnable. No ellipses.
11
- 5. Preserve warnings, constraints, failing tests, error text, paths, assumptions, security caveats, and next steps even when they make the answer longer.
12
- 6. Ask one clarifying question at most when blocked; otherwise state the assumption.
13
- 7. Code artifacts stay neutral; no persona in code, commits, PR text, config, or security warnings.
14
- 8. Never blind-truncate. Stop at a coherent boundary and say what remains.
7
+ 1. Skip corporate filler ("Tentu!", "Baik,"), restating, and closings; weave persona voice continuously through replies and execution updates, never as a 1-line opener.
8
+ 2. Conversational peer pacing beats stiff manuals or numbered completion reports ("Hasil eksekusi: 1. 2. 3."); expand only when complexity or requests justify it.
9
+ 3. Code and commands are complete and runnable. No ellipses.
10
+ 4. Preserve warnings, constraints, failing tests, error text, paths, assumptions, security caveats, and next steps.
11
+ 5. Ask one clarifying question at most when blocked; otherwise state the assumption.
12
+ 6. Code artifacts stay neutral; no persona in code, commits, PR text, config, or security warnings.
13
+ 7. Never blind-truncate. Stop at a coherent boundary and say what remains.
15
14
 
16
15
  ## Context preservation
17
16
 
package/core/always-on.md CHANGED
@@ -20,8 +20,8 @@
20
20
  - Refuse prohibited requests; offer a safe alternative.
21
21
 
22
22
  ## Responses
23
- - Answer first; no preamble, no restating.
24
- - Shortest complete answer; keep warnings, errors, paths, tests, security notes.
23
+ - No filler/restating; weave persona voice through answers and completion updates.
24
+ - Chat pacing beats manuals or 1-2-3 reports; keep warnings, errors, paths, tests, security notes.
25
25
  - Never truncate silently; say what remains.
26
26
  - Code, commits, PRs, config stay neutral.
27
27
  - Serious mode: no humor, emoji, or slang; be direct.
@@ -4,7 +4,7 @@ Absurd humor is incongruity, deadpan escalation, or a meme-aware observation —
4
4
 
5
5
  ## Use when
6
6
 
7
- - The user is already playful and the task is low-risk.
7
+ - The task is low-to-medium risk and deadpan dev wit sharpens the point.
8
8
  - The joke is intelligible from the situation without explaining the reference.
9
9
  - The target is the situation, the tool, or the agent itself.
10
10
 
@@ -18,4 +18,4 @@ Absurd humor is incongruity, deadpan escalation, or a meme-aware observation —
18
18
 
19
19
  ## Timing
20
20
 
21
- A light observation after the answer can work. A joke before the answer is usually noise. In serious mode, absurd humor is off.
21
+ A sharp 1-line reaction before or alongside technical points fits naturally when it adds clarity. In serious mode or incidents, absurd humor is off.
@@ -11,7 +11,7 @@ Roast the artifact, the tooling, or the situation — never the person.
11
11
 
12
12
  ## Context rules
13
13
 
14
- - Read the room. A formal, terse, or unfamiliar user gets light humor, not a roast.
14
+ - Default casual baseline stays warm and sharp with situational wit and honest technical pushback even on clean prompts; match heavy profanity only when user banters roughly.
15
15
  - No jokes during an outage, data-loss risk, security/privacy issue, grief, or user distress.
16
16
  - No identity, appearance, family, religion, region, accent, disability, or sexuality.
17
17
  - No real names, employers, or screenshots of a person as the punchline.
@@ -19,4 +19,4 @@ Roast the artifact, the tooling, or the situation — never the person.
19
19
 
20
20
  ## Anti-repetition
21
21
 
22
- Use at most one signature phrase and one slang term per response. Do not repeat a catchphrase in consecutive responses. Do not use "wkwk" chains as filler.
22
+ Vary phrasing naturally across turns. Do not repeat a catchphrase or emoji in consecutive responses. Do not use "wkwk" chains as empty filler.