brainstorming
/brainstorming
Burzy pomysły na zmianę w kodzie, sprawdzając ryzyko, właścicieli i długi techniczne obszaru, zanim powstanie projekt.
- Autor
- artk0de ↗
- Repozytorium
- tearags-mcp ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 10.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 10.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Burzy pomysły na zmianę w kodzie, sprawdzając ryzyko, właścicieli i długi techniczne obszaru, zanim powstanie projekt.
Wyzwalacze z opisu
Oryginalny opis (EN)
Brainstorm code change seeing risk, ownership, tech-debt signals from tea-rags first — creative exploration grounded in actual state of affected area. Triggers on "brainstorm X" (incl. "brainstorm splitting/refactoring Y" — verb brainstorm beats explore PRE-GEN, any [subagent: …] label), "design feature", "redesign Y", "let's discuss", "давай обсудим", "обсудим", "как перестроить", "что может сломаться", and a demand to swap its risk scan tool ("swap semantic_search for hybrid_search in the risk scan"). NOT research before refactor without a brainstorm/discuss verb (→ explore PRE-GEN), NOT tri
Jak działa
- Iron Rule
- Step 1 — Extract target area
- Step 2 — Codegraph decision, then ONE parallel block (A–E on / A–C off)
- Step 3 — Build enrichment block
- Step 4 — Invoke superpowers:brainstorming
- Red Flags — STOP, restart from Step 2
Podgląd SKILL.md
---
name: brainstorming
description: >-
Brainstorm code change seeing risk, ownership, tech-debt signals from tea-rags
first — creative exploration grounded in actual state of affected area.
Triggers on "brainstorm X" (incl. "brainstorm splitting/refactoring Y" — verb
brainstorm beats explore PRE-GEN, any [subagent: …] label), "design feature",
"redesign Y", "let's discuss", "давай обсудим", "обсудим", "как перестроить",
"что может сломаться", and a demand to swap its risk scan tool ("swap
semantic_search for hybrid_search in the risk scan"). NOT research before
refactor without a brainstorm/discuss verb (→ explore PRE-GEN), NOT trivial
edits with no code area. Wraps superpowers:brainstorming with tea-rags risk
signals.
codex-description: >-
Brainstorm code change grounded in tea-rags risk, ownership, tech-debt
signals. Verb brainstorm/discuss beats explore PRE-GEN ("brainstorm splitting
Y"). Triggers "brainstorm X", "design feature", "redesign Y", "let's discuss",
"давай обсудим", "обсудим", "как перестроить", "что может сломаться". NOT
research w/o brainstorm verb, NOT "what breaks if I change <symbol>"
(explore), NOT trivial edits.
---
# dinopowers: brainstorming
Wrapper over `superpowers:brainstorming`. Exploration starts from real git
signals (hotspots, owners, legacy debt) of target area — not assumptions.
## Iron Rule
**tea-rags enrichment MUST run BEFORE `Skill(superpowers:brainstorming)`** when
target area identifiable. Right presets (`hotspots`, `ownership`, `techDebt`) +
right params + right order = wrapper's value.
No code area identifiable (pure conceptual) → skip enrichment, invoke
`superpowers:brainstorming` directly, say so explicitly. Never fabricate area.
**User explicitly asks to read named files / find callers of a symbol** → those
reads are the USER's task, not wrapper enrichment: planned in the SAME answer as
separate calls AFTER the Step 2 block returns — never inside the parallel block,
never deferred to a later turn. One named file →
`find_symbol relativePath=<file>`; several → ONE `find_symbol symbols=[<files>]`
(core Invariants), never one call per file;
`hybrid_search metaOnly:true query=<symbol>` for callers (graph tools on →
`get_callers`). Print:
`user reads: find_symbol / hybrid_search metaOnly — separate calls AFTER the Step 2 block returns, same answer, not deferred`.
**Otherwise Step 2 block = this wrapper's FIRST tool calls; the wrapper adds no
`find_symbol` of its own** — not before Step 2 (a user-named area / dir becomes
`pathPattern`), not after (enrichment rows are metaOnly signals, handed to
brainstorming as-is; the "after search → `find_symbol`" depth rule does not fire
on them). Print:
`first calls: Step 2 block; wrapper adds no find_symbol — none after it either, not "if rows need inspection"; rows go to brainstorming as-is`.
User asks to SEE file content → Step 2 "Do NOT pass" content line.
**User demands another scan tool** (swap `semantic_search` for `hybrid_search`
in the risk scan, "more reliable") → stay in this wrapper, refuse, print
verbatim, then Step 1 → 2 → 4 over the named area:
```
hybrid_search refused — same rerank, but its BM25 leg re-ranks by query-token overlap; risk signals want semantic + signals. Running semantic_search with rerank presets hotspots / ownership / techDebt (metaOnly).
```
Never reroute that demand to `risk-assessment` / `rank_chunks` /
`analytics-rerank`.
**Chaining rule:** [CHAINING.md](../../CHAINING.md) — every onward
`superpowers:Y` → `dinopowers:Y`. NEVER bypass wrapper.
**Requires superpowers:** `superpowers:brainstorming` — availability per
[CHAINING.md](../../CHAINING.md) "Missing superpowers" steps 1–3. Not in the
skills list ≠ unavailable (the list is truncated): run
`ls $CODEX_HOME/plugins/cache/*/superpowers/*/skills/brainstorming/SKILL.md` as
written first — a hand-built path failing proves nothing; only that glob
matching nothing → stop and ask the user to install superpowers
(https://github.com/obra/superpowers). NEVER present this enrichment alone as
the full workflow.
**Freshness:** [FRESHNESS.md](../../FRESHNESS.md) (owner
`tea-rags://guide/freshness`). Lag ≠ unusable — never Bash `cat`/`grep`/`Read`
over code; `treeState` row → `find_symbol`.
## Step 1 — Extract target area
| Element | Example |
| ---------------------------------------------- | ------------------------------------------------------- |
| **Area** — path or domain | `src/core/domains/ingest/**`, `auth/`, "rerank presets" |
| **Change type** — what the brainstorm is about | "new feature", "refactor", "bug-fix approach" |
| **Scope hint** — narrower filter if mentioned | file name, class name, symbol |
Disqualifier first: user rules files out ("no specific files", "без конкретных
файлов", "conceptually") → no area, a subsystem word in the request ("cache
layer design") notwithstanding — never glob it; Step 4 without enrichment.
Directory scope → `pathPattern="<dir>/**"` (`src/app/widgets/` →
`src/app/widgets/**`); a bare `<dir>/` matches no file. Print it:
`dir "<dir>/" → pathPattern="<dir>/**"`.
Named path new / not in git yet → `pathPattern` = that path exactly (dir →
`<dir>/**`), Step 2 calls still RUN (new path ≠ no area). NEVER widen to a
parent dir to find signals.
Compose `pathPattern` (e.g. `**/auth/**`,
`src/core/domains/ingest/**/chunker/**`) + `query` (one sentence, change
intent). Area named by a word, no directory given ("the auth flow", "cache
layer") → glob of that word, `**/<word>/**` (`**/auth/**`, `**/cache/**`) — set
on EVERY enrichment call; pathPattern unset while an area is named = un-scoped
enrichment, wrong. Print it:
```
area "<named area>" → pathPattern="**/<word>/**"
```
No area mentioned/derivable → Step 4, empty enrichment. Report "no area
identifiable — proceeding without tea-rags enrichment".
## Step 2 — Codegraph decision, then ONE parallel block (A–E on / A–C off)
**2.0 — decide codegraph state FIRST, state it in output before the block:**
graph tools (`get_callers`, `trace_path`, …) in your tool list → print the
`codegraph on` line of the block below; absent → its `codegraph off` line. No
probe call. No stated decision = Step 2 not started.
Plan-only with the state not given in the prompt → print BOTH lines, each branch
complete (the off line is never implied by the on line):
```
codegraph on (graph tools in my list) → A–E in one block: A hotspots, B ownership, C techDebt, D architecturalHub (semantic_search, metaOnly, pathPattern=<Step 1>) + E get_naming_lexicon(concept, language, pathPattern=<Step 1>)
codegraph off → A–C in one block: A hotspots, B ownership, C techDebt (semantic_search, metaOnly, pathPattern=<Step 1>); backbone + vocabulary not assessed
```
Print these lines verbatim — also when entered from another wrapper (plan step,
subagent). "follow the wrapper's enrichment order" / "run Step 2" without the
calls + presets named = Step 2 not stated.
Presets FIXED: A hotspots, B ownership, C techDebt — area type never swaps one
(security / auth / payments area → still `techDebt`, NEVER `securityAudit` or
another preset in its place).
Then ALL calls **in parallel** (one tool block — D + E in the SAME block as A–C,
never appended after): A–D `mcp__tea-rags__semantic_search`, each
`metaOnly: true` (signals, not content); E `get_naming_lexicon`:
| Call | `rerank` preset | Why |
| -------------- | ----------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
| A — Hotspots | `"hotspots"` | Bug-prone zones the brainstorm must account for (high churn + recency + bugFix) |
| B — Ownership | `"ownership"` | Knowledge silos — single-owner files that need reviewer pairing |
| C — Tech debt | `"techDebt"` | Legacy zones where "just refactor it" hides cost (age + churn + bugFix) |
| D — Backbone | `"architecturalHub"` | **Codegraph only** — structural hubs (high `fanIn` / `isHub`) the design must respect; a change to a backbone has wide blast radius |
| E — Vocabulary | `get_naming_lexicon` (not a search) | **Codegraph only** — words the project already uses for the area; SAME parallel block as A–D |
Call D ONLY when codegraph on = context listing `codegraph.symbols`, or
`get_naming_lexicon` available (explicit, no further check), or graph tools
(`get_callers`, `trace_path`, …) in tool list; absent = off. Off → A/B/C only,
note backbone not assessed (NEVER pass similarity-ranked list off as "the
hubs"). → `tea-rags://guide/navigation` "Graph navigation" — for the on/off read
only; its after-search `find_symbol` depth step does not apply to enrichment
rows.
**Call E — vocabulary (codegraph only, same parallel block — MANDATORY when
codegraph on; A–D without E = incomplete Step 2):**
`get_naming_lexicon(concept=<Step 1 query>, language, pathPattern)` →
`concept.terms` = words project already uses for area. `concept` = change intent
sentence, NEVER a draft symbol name. `language` = area's language from its path
/ project stack; unsure → still run E, never skip it over a language-capability
doubt. Output → two block lines, both printed: "Project vocabulary" from
`concept.terms`, and "Names check (handoff)": names the design proposes for new
symbols go through `get_naming_lexicon(names=[…])` before being fixed.
Exact parameters per call:
```
project: <alias> ← `tea-rags://guide/freshness` "Addressing the Codebase"
path: <your working directory> ← the tree (worktree / branch checkout)
query: <intent sentence from Step 1>
pathPattern: <pathPattern from Step 1>
rerank: "hotspots" | "ownership" | "techDebt" | "architecturalHub"
limit: 10
metaOnly: true
```
(`"architecturalHub"` = Call D, codegraph-gated per the note above.)
Do NOT substitute:
| Wrong tool | Why wrong |
| ------------------------------ | ---------------------------------------------------------------------------------------------------- |
| `mcp__tea-rags__hybrid_search` | Same rerank, but its BM25 leg re-ranks by query-token overlap — signal scan wants semantic + signals |
| `mcp__tea-rags__find_similar` | Requires an existing symbolId; brainstorm targets an area, not a known symbol |
| `mcp__tea-rags__find_symbol` | Returns one symbol; brainstorming needs area-wide risk distribution |
| Running the calls sequentially | Wastes time; they are independent — issue them in one parallel block |
Do NOT pass:
- `metaOnly: false` — content unneeded, payload bloats. No "show / read the
content" in the prompt → no content call, conditional ones included. Prompt
asks to SEE content → enrichment stays `metaOnly: true`; content = separate
call AFTER the Step 2 block returns (same answer, never in the block, never
deferred): `find_symbol relativePath=<row.relativePath>` (outline → drill
ids); several rows → ONE `find_symbol symbols=[<row paths or ids>]`
- custom `rerank` weights instead of named presets — loses calibrated semantics
- `filter` on `git.ageDays` / `git.commitCount` — presets encode them; filters
shrink set, hide debt
All risk calls 0 results (new / unindexed area) → report "no git-enriched
results for pathPattern `<X>` — area is new or excluded from indexing", Step 4
WITHOUT block: no Step 3 block, no `[planned; not run]` rows, never "pass the
enrichment". No retry: no wider `pathPattern`, no other tool. NEVER fabricate.
Plan-only, path new / not in git → calls are NOT skipped: list the A–C calls (+
D/E when on) with pathPattern = that path, then the outcome of their result:
```
A hotspots, B ownership, C techDebt (semantic_search, metaOnly, pathPattern=<new path>) → when they return 0 rows: "no git-enriched results — area is new"; superpowers:brainstorming invoked WITHOUT enrichment block
```
NEVER "skip the enrichment calls" / "no tea-rags call" because the path is new —
0 rows is the calls' result, not a reason to not issue them.
## Step 3 — Build enrichment block
Compact block (max ~15 lines):
```
### tea-rags enrichment for <pathPattern>
**Hotspots (bug-prone):**
- <relativePath>:<startLine>-<endLine> — bugFixRate <N>, commitCount <N>, ageDays <N>
- ... (top 3)
**Knowledge silos (single live-line owner):**
- <relativePath> — <blameDominantAuthor> (<blameDominantAuthorPct>%), <commitCount> commits
- ... (top 3)
**Tech debt (legacy + churn):**
- <relativePath> — ageDays <N>, commitCount <N>, relativeChurn <X>
- ... (top 3)
**Project vocabulary (Call E):** <term> (<holder>), ... (top 5)
**Names check (handoff):** names the design proposes for new symbols go through get_naming_lexicon(names=[{name, kind}], language) before being fixed — MISFIT → take suggestion; NEW_TERM → vocabulary term if same meaning, else justify the new word.
```
Codegraph on → both last lines printed verbatim, plan-only answer included
(`[planned; not run]` covers the terms, never the Names check line).
Cap 3 per section. File in several sections → mark (`→ also in Hotspots`) —
overlap = strong signal. No raw JSON; `relativePath` + 2-3 key signals per
entry. Values: `rankingOverlay.{file,chunk}.<field>` (preset-surfaced, labelled)
— the only signal source: a metaOnly payload is locator-only, no raw `git.*`; a
signal the preset does not surface → leave it out. File-level presets
(ownership, architecturalHub) → no startLine.
**Done-check before Step 4 (codegraph on) — state all three, else Step 2/3
incomplete:** (1) E issued in the same parallel block as A–D; (2) block carries
"Project vocabulary" line from `concept.terms`; (3) block carries the "Names
check (handoff)" line verbatim.
## Step 4 — Invoke superpowers:brainstorming
Invoke `Skill` tool with `superpowers:brainstorming`. Prepend enrichment block
from Step 3 as context. Phrase handoff as:
> "Before exploring, note these risk signals in the target area: …<block>… Use
> them to pressure-test ideas — especially around the hotspot files. Apply the
> block's Names check line to every name the design proposes for a new symbol.
>
> Chaining rule reminder: if your cycle would next invoke a `superpowers:Y`
> skill (writing-plans, test-driven-development, etc.), invoke the
> `dinopowers:Y` wrapper instead when one exists — see the Chaining rule section
> above."
`superpowers:brainstorming` runs full cycle — wrapper grounds, never replaces.
## Red Flags — STOP, restart from Step 2
- "I know the hotspots here" → run Step 2; memory stale
- "3 calls slow, do one" → presets = different lenses; one loses two
- Custom weights / `rerank: "relevance"` instead of risk presets → redo
- `hybrid_search` / `find_similar` / `find_symbol` substituted → redo with
`semantic_search` + preset
- Wrapper-own `find_symbol` before or after Step 2 (named dir, result paths, doc
TOC) → drop it; Step 2 block first, brainstorming next. User-asked reads of
named files / callers stay (see Iron Rule)
- Scan-tool demand rerouted to `risk-assessment` / `rank_chunks` → back here,
print the refusal line, `semantic_search` presets
- Calls sequential → reissue parallel
- Block issued without stated codegraph decision, or graph tools in list but
only A–C → add D + E in the same block, restate
- `superpowers:brainstorming` first, tea-rags after as "validation" → wrong
order, restart
- `metaOnly: false` "to see content" → restart with `metaOnly: true`
- Raw `results[]` JSON in handoff → extract 3 entries per lens
- User named area but skipped as "abstract" → enrich it
- Inner cycle chains raw `superpowers:Y` → intercept, invoke `dinopowers:Y`
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · artk0de/tearags-mcp | 100 | MIT |
| jjuraszek/pi-gauntlet | 100 | MIT |
| GanyuanRan/Aegis | 100 | MIT |
| Montinou/triqual | 100 | MIT |
| guanyang/antigravity-skills | 100 | MIT |
| osmontero/opencode-skills | 100 | MIT |
| Sanit-mann/The-Insectoids | 100 | MIT |
| coco-research/coco | 100 | MIT |
| ILIV007/prime-skill | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki bd93c807d10e (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100dino-brainstorming
Prowadzi burzę mózgów o zmianie w kodzie, opierając się na sygnałach ryzyka, właścicielstwa i długu technicznego z narzędzia tea-rags. - 100writing-plans
Pisze plan zmiany, a listę plików wzbogaca o importy, zmienność, właścicieli i dane o poprawkach z repozytorium. - 100phx-brainstorm
Prowadzi burzę mózgów o funkcjach Elixir i Phoenix: porównuje podejścia i zbiera wymagania, zanim powstanie plan. - 90trove-brainstorm
Wymusza burzę mózgów przed nowymi funkcjami, zmianami UI, refaktoryzacją lub produktem, zanim powstanie zatwierdzony projekt. - 100brainstorm
Prowadzi dialog przed pracą kreatywną lub istotnymi zmianami: pytania pojedynczo, eksploracja podejść i stopniowa walidacja zamieniają pomysły w projekty funkcji i decyzje architektoniczne. - 90brainstorm
Prowadzi sesję burzy mózgów, która porównuje opcje i porządkuje decyzję, przydatną przy pytaniach o alternatywy, wady i zalety oraz wybór.