Skip to content
Agentmart
Submit a tool

explore

Exploration through foggy or oversized work.

Skill by marcuskrogh

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/marcuskrogh-explore

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Exploration through foggy or oversized work. Charts ROADMAP.md plus a Story of sequenced route Tasks (one delivery unit by default). Research, model, and sandbox artifacts stay on the delivery branch — no separate PRs. Use when the destination is felt but the route is not yet visible.

Use cases: research

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/explore

What's inside

Skill name
explore
Description
Exploration through foggy or oversized work. Charts ROADMAP.md plus a Story of sequenced route Tasks (one delivery unit by default). Research, model, and sandbox artifacts stay on the delivery branch — no separate PRs. Use when the destination is felt but the route is not yet visible.
License
MIT
Instructions
152 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Explore

Applies [CONCEPT_ALIGNMENT](../concepts/CONCEPT_ALIGNMENT.md) and
[CONCEPT_SKILL](../concepts/CONCEPT_SKILL.md) to **wayfinding** — charting a
course through foggy work. Explore charts. The explore workflow walks.

**On invoke:** read [../concepts/CONCEPT_ALIGNMENT.md](../concepts/CONCEPT_ALIGNMENT.md),
[CONCEPT_SKILL](../concepts/CONCEPT_SKILL.md),
and [../workflow/SKILL.md](../workflow/SKILL.md).

## Leading words

- **delivery unit** — one define-typed Task (and its PR) that owns research/model/sandbox artifacts through ship when they share a build
- **finding docs** — `RESEARCH.md` / `MODEL.md` committed on the delivery branch for later skills; never their own PR
- **supportive-only** — explore route Task that does not itself implement or ship; its artifacts sit on the delivery branch

**Chart the route.** Creating a research/model/sandbox/define Task completes a map step.
The explore workflow reads the frontier type.

## Extensions

| Slot | This skill |
|------|------------|
| **Subject** | Vague, foggy, or oversized work |
| **Probes** | Destination (1–2 lines); why it matters; competing framings that change the whole map; fog vs ticket; route step kinds; sequence/deps; out of scope |
| **Stop condition** | Destination named, visible frontier ticketed (with sequence/deps), remaining uncertainty recorded as fog or out of scope |
| **Alignment artifact** | `ROADMAP.md` (path from WORKSPACE) — the map |
| **Readiness prompt** | "Does this map capture the destination and the next steps through the fog?" |
| **Opening** | Thin: "What are we trying to find our way to?" Rich: highest-value fog (destination, framing, or first takeable step). Existing map → [Continue the map](#continue-the-map) |
| **Scope guard** | Chart only — no destination implementation; no fake-settled requirements for later skills; tracker writes after approval are charting, not walking; no map-only open PR |

**Divergence here** changes destination, route shape, or Task existence/deps.
Choices that only change how a later skill answers belong on that Task — park a
one-line fog/Task pointer.

## Fog of war

**Fog or ticket?** Can you state the step precisely enough to act on *now* — not
whether you can answer it now.

| Put it here | When |
|-------------|------|
| **Route Task** | Sharp enough to ticket as type define (preferred), research, model, sandbox, or task — even if blocked |
| **Not yet specified** (fog) | Sensed but not ticket-sized — do not pre-slice into fake Tasks |
| **Out of scope** | Beyond this destination — never graduates unless destination is redrawn |

Resolving a route Task should clear fog ahead. Re-invoking `/explore` **graduates**
what is now sharp into fresh Tasks.

## Route Task types

| Type | Role |
|------|------|
| **research** | Evidence finding docs the route waits on |
| **model** | Math finding docs before (or with) definition |
| **sandbox** | Isolated inspect-loop for a contained element before production implement |
| **define** | User-agent alignment on particulars for a buildable slice |
| **task** | Other unblocker that earns its place by unblocking a later decision |

**One delivery unit by default.** Prefer a **single define-typed route Task**
when research / model / sandbox / define feed the same eventual build — run those
steps on that Task so artifacts land on the same branch define and implement
will use. Prefer **separate route Tasks with dependencies** only
when sequence truly needs independent tickets (parallel owners, research that
may kill the destination, or blockers owned elsewhere).

Wire **Blocked by** after Tasks have keys. A Task is **unblocked** when every
blocker is **Done** (or waived). The frontier is the unblocked row with the
lowest Order. The explore workflow maps its type to a skill.

**Finding docs, not separate PRs.** Research and model only produce finding
documentation on the delivery branch. Sandbox produces an isolation tree and
`SANDBOX.md` the same way. They never open their own PRs. Explore does not open
a map-only PR. The only open delivery PR is the delivery head once a definition
skill opens it. Supportive-only route Tasks go **Done** after their docs (or
isolation tree) are on the downstream delivery branch.

## Artifact

`ROADMAP.md` is an **index**; detail lives on tickets.

```markdown
# Roadmap: [title]

## Destination
<one or two lines>

## Notes
<domain hints; standing preferences>

## Route
| Order | Task | Type | Blocked by | Status | Issue |
|-------|------|------|------------|--------|-------|
| 1 | <title> | research / model / sandbox / define / task | — | To Do | <KEY> |

## Cleared so far
- [<title>](link) — <one-line gist>

## Not yet specified
- …

## Out of scope
- …

## Tracker
- Provider: …
- Story (map): <KEY>
- Tasks: <KEY>, …
```

Keep Destination short. **Not yet specified** stays non-empty whenever honest fog remains.

## Steps

### Chart the map

1. **Align** — Follow CONCEPT_ALIGNMENT with the extensions above. Done when the destination and map stop condition hold.
2. **Classify the route** — Map the visible frontier breadth-first into typed Tasks, fog, and out of scope. Done when every visible item has exactly one classification and dependencies are known.
3. **Handle a clear direct path** — If the work is already small and fog-free, record a one-row route of type **define** instead of manufacturing extra Tasks. Done when that row is the frontier.

From marcuskrogh/skills, MIT.

More from marcuskrogh/skills

Signals

Stars
1
Forks
1
Open issues
0
Last commit (this folder)
Sep 27, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from marcuskrogh

Cite this page

explore skill. Agent Mart. https://agentmart.dev/skills/marcuskrogh-explore (last updated 2026-10-24)