| name | chunked-discovery |
| description | For broad domains ("AI", "fintech", "climate"), segment into sub-niches first, then drill down per segment using discover-by-domain. Writes a parent index plus per-segment notes. Use when discover-by-domain bails out. |
| disable-model-invocation | false |
| allowed-tools | Read, Write, WebSearch, WebFetch, Bash(mkdir *), Bash(test *) |
Chunked Discovery
Broad domains can't be enumerated honestly in one pass — the long tail goes off-topic before you get coverage. This skill segments first, then drills.
Inputs
$ARGUMENTS:
- Required:
<broad-domain> — e.g. "AI", "fintech", "climate tech".
- Optional:
--segments=<n> (default 6) — number of sub-niches to derive.
- Optional:
--depth=<count> (default 10) — companies per segment.
- Optional:
--auto-drill — run discover-by-domain on each segment automatically. Otherwise, write the index and stop.
- Optional:
--refresh.
Examples
/career:discover chunked "AI"
/career:discover chunked "fintech" --segments=8 --auto-drill
/career:discover chunked "climate tech" --depth=15
Procedure
1. Resolve config + paths
Output index: ${WORKING_FOLDER}/domain-notes/chunked-${broad-slug}.md. Per-segment notes: ${WORKING_FOLDER}/domain-notes/${segment-slug}.md (same shape as discover-by-domain output).
2. Derive segments
WebSearch for "landscape", "market map", VC theses, and analyst reports on the broad domain. Synthesize --segments non-overlapping sub-niches. For each:
- Name (kebab-case slug + display name).
- One-line definition.
- Distinguishing axis vs adjacent segments (so the user can spot overlap).
Present the segment list to the user and ask for confirmation / edits before drilling.
3. Drill (if --auto-drill)
For each confirmed segment, invoke discover-by-domain with:
<domain> = segment display name
--slug = segment slug
--n = --depth
Skip segments where a fresh note already exists (< 30 days old) unless --refresh.
4. Write the index
# Chunked discovery — {{broad_domain}}
> Surveyed: {{date}}
> Segments: {{n}}
> Drilled: {{true|false}}
## Segments
| slug | name | one-liner | drilled note |
| --- | --- | --- | --- |
| ... | ... | ... | domain-notes/<segment-slug>.md (if drilled) |
## How segments differ
- {{seg-a}} vs {{seg-b}}: ...
## Suggested drill order
1. {{slug}} — reason
2. ...
## Sources
1. ...
5. Print summary
chunked index: domain-notes/chunked-<broad-slug>.md
segments: <n>
drilled: <n> (if --auto-drill)
skipped (recent): <n>
next:
if not drilled → /career:discover by-domain "<segment>" for each
if drilled → /career:suggest companies --domain=<slug> per segment
Guardrails
- Segments must be non-overlapping. If two segments ≥30% overlap, merge or re-cut.
- Don't fabricate segments. If the broad domain is genuinely under-segmented (e.g. a frontier with only one shape), say so and recommend
discover-by-domain directly.
--auto-drill is bulk WebSearch + WebFetch — warn the user about cost before running with --segments=8 --depth=20.
Failure modes
- Domain still too broad even after segmentation ("technology") → bail; ask the user to narrow.
- Drill failure on a segment → mark that segment as
## Status: partial in its note; continue with others.