One workflow: before you formally commit, clarify three things at the same time—collision risk, the evidence base in the quiet zone, and potential challenges. These three steps are not simply chained together—the results of collision detection affect the search direction of the latter two steps: the searches for support and challenges focus on thequiet zone, rather than the full picture of the entire idea. This makes bear-propose more valuable than running three skills separately.
First read references/sci-cli.md—CLI detection steps, usage, and iron rules.
Then read references/output-system.md—HTML shell specification and design tokens.
Steps (fixed order)
Step 0 — Confirm the CLI is available
Run sci --version. If not installed, give the installation command and stop; if there is an authentication error, prompt for sci init and stop.
Step 1 — Compress the idea and run collision detection
Distill the idea into one sentence. Search from 5–6 orthogonal angles (literal phrasing / method-centered / problem-centered / conclusion-centered / adjacent fields / preemptor titles); use ultra_lowfor broad angles and use low to rerun angles that yielded results, merge and deduplicate, and classify into four tiers (direct collision / method twin / problem twin / neighbor).
Tell the user that this run will perform about 15–20 searches, then start directly without waiting for confirmation.
Step 2 — Identify the quiet zone Based on the collision results, clarify which angles are crowded and which are quiet. The quiet zone is the focus of subsequent searches—finding support in the quiet zone is what makes a truly meaningful literature base.
Step 3 — Quiet-zone supporting-evidence search
For 1–2 core angles in the quiet zone, run sci search（--mode low，--limit 15), to find literature that supports this idea being meaningful in the quiet zone. Screen according to four-level criteria. If the support signal in the quiet zone is very weak, say so honestly—this means few people have indeed worked on this direction, but it may also be because doing so has no value.
Step 4 — Potential-challenge search
For the core claims in the quiet zone, run opposing-literature searches (focusing on two categories, direct contradiction and methodological criticism, using one low search for each), to identify the challenges that need to be thought through in advance. There is no need to cover all five categories; search only the two categories most relevant to project launch.
Step 5 — Comprehensive judgment Based on the search results of the three phases, write 3–5 sentences describing the pre-launch evidence landscape: the collision risk level, the strength of the support signal in the quiet zone, and the types of challenges that most need to be addressed in advance in the research design. Do not make a final judgment on "whether this idea is worth doing".
Outputs
After the searches end, create in the current directory the {topic-slug}/ subfolder and output four things:
- Terminal —directly print the comprehensive summary: collision risk level + main angles of the quiet zone + challenges that most need to be thought through in advance
report.md—stable YAML front matter + comprehensive judgment + collision tiers + quiet-zone support + potential challenges + search transparencyreport.html—comprehensive report, see the HTML specification belowreferences.bib— Merge all queries'.bib, deduplicate, and keep only the references cited in the report
HTML report specification
Accent color #d97706(amber, conveying the prudence of "before deciding", distinct from bear-scoop's blue and bear-counter's red).
First screen: overall judgment panel
.verdict block in three horizontal segments:
- Left: collision risk summary (level + most recent collision candidates, blue accent)
- Center: evidence landscape description in 3–5 sentences (amber accent, core position)
- Right: most critical challenge (type + one-sentence core challenge content, red accent)
.metric-strip: direct collision count / effective supporting literature count in the quiet zone / high-threat challenge count / key points requiring advance response
Detailed content in three phases (tabs, three tabs)
Tab "Collision Detection": SVG four-layer concentric circle radar (same specification as bear-scoop), with a four-layer literature list below. The radar hover tooltip is implemented with JS mousemove.
Tab "Quiet Zone Support": evidence ladder (direct support / partial support / indirect relevance), including a quiet zone explanation—why these papers are valuable to the project (they support an angle others have not yet pursued). Literature cards include DOI links and collapsible Chinese academic abstracts.
Tab "Potential Challenges": displayed by two types of challenge directions (direct contradiction / methodological criticism), each card includes threat level + 回应思路：... + DOI link + collapsible Chinese academic abstract.
The most important cards in the three tabs (most recent collision, strongest quiet zone support, highest-threat challenge) get the .card-primary class, and abstracts are expanded by default.
Search transparency(bottom): lists all query terms and the search distribution for each phase, explaining how the quiet zone is determined and how the latter two steps focus.
Honest boundaries
The quiet zone is a relative judgment based on this collision detection and does not mean no one is working on it—behind paywalls, recent preprints, and Chinese-language literature may all be missed. A weak quiet zone support signal may mean the direction is novel, or it may mean predecessors tried it but it had no publication value. The report presents what was retrieved and does not make a judgment about "whether it is worth doing".
Gotchas
- Step 3 and Step 4 must focus on the quiet zone, do not conduct broad support and opposition searches for the entire idea—that would be no different from running the three skills separately
- A weak quiet zone support signal is not a bad thing, present it truthfully in the report, and do not lower the screening criteria just to "look supported"
- The three tabs use their respective original skill accent colors(blue/green/red) as accents, and the overall report uses amber as the framework color; do not confuse them
- If collision detection finds more than 5 direct collisions, before Step 3 begins, first notify the user: the quiet zone is very narrow, and subsequent searches will focus on more segmented angles; the reference value of this report depends on whether the user is willing to continue adjusting the idea in a segmented direction