Atlas / Skills / igorkravcenko / snowshoe

snowshoeSAFE

skills/igorkravcenko/snowshoe

Don't let your agents outrun your understanding. Keep your footing. A git-native personal comprehension map of a repository — tracks what goes stale after pull. Catch up after pull.

Verdict
SAFE
Grade
B
Trust score
89 /100
Version
—
Hosts
1 documented
License
Apache-2.0
Stars
1
01

Overview

Don't let your agents outrun your understanding. Keep your footing. A git-native personal comprehension map of a repository — tracks what goes stale after pull. Catch up after pull.

Read from source at commit c1fbd5a3e704OBSERVED · 2026-10-10
02

Install

Commands as the repository documents them. They are shown, not run.

npm install -g @igorkravcenko/snowshoe
03

Host compatibility

What the documentation claims. We have not run a compatibility test.

HostStatusNotes
cursormentioned
04

What it tells the agent

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: snowshoe
description: >-
  Snowshoe personal comprehension map of a repository. After pull or catch-up: drain via work next
  (init / routine refresh / routine advance, then complete claimed steps).
  In the map PTY or when studying the map (SNOWSHOE_MODE=learn): talk about
  the tree using map view / map status. Combined mode (map UI / UI terminal open while
  draining): read both drain.md and learn.md. PATH binary snowshoe; install.md
  only if the binary is missing.
---

# Snowshoe skill

Assume **`snowshoe` is on PATH**. Always pass `--json` and parse the result.

The install binary is **`snowshoe`** (call this).

If this working repo is missing the skill files under the harness skills dir, run
`snowshoe skill install --json --skills-path <harness-skills-dir>` (e.g. `.cursor/skills`
for Cursor). Opt-in; not part of `init`. Prefer `snowshoe help --json` over guessing flags.

Drive the CLI. Do not write the ledger yourself. Do not start `map serve` unless the human asked.

If **this turn** they asked to mark, expand, enrich, fix, detail, or cancel named node(s) (or “this node”): `snowshoe map mark --json --slug <slug> --kind detail` (or `expand` / `enrich` / `fix`; `map detail mark` queues `detail`). Do not mark a tour of the tree. Then continue the drain/learn branch.

Read sibling files after the checks below. Default: **one** of `drain.md` or `learn.md`. **Combined** mode (below) may read **both**. Do not open `install.md` unless `snowshoe` is missing.

## 1. Binary

If `command -v snowshoe` fails, read **`install.md`** in this same folder — only then.

After following it, run `command -v snowshoe` again:

- Still missing → **STOP**. The binary often needs a **new shell** after global install or `bun link`. Tell the human how to re-check; do not start drain or learn.
- Now present **and** this turn already classified drain, learn, or combined → go to **Intent** and open that file.
- Now present **but** they only asked to install, or intent is still unclear → ask once: drain, learn, combined, or stop.

## 2. Intent

Classify in this order:

1. The **user prompt that invoked this skill**.
2. Nearby user text **this turn only** (not the rest of the thread).
3. If text is still empty: `SNOWSHOE_MODE=learn` → learn. Other/empty mode does not override an explicit drain prompt.
4. Else **drain**.

| Drain-adjacent | Learn | Combined |
|---|---|---|
| after pull, catch up, drain, `work next`, marked nodes, complete queue | study / explain the map, “what is this node”, look at the map, map PTY agent | map UI open / UI terminal + keep draining while studying/learning in the **same** session |
| | `SNOWSHOE_MODE=learn` when (1)–(2) are empty | same-session foreground learn/map conversation **and** background `work next --wait` |

Explicit drain in the prompt beats env **unless** combined signals apply.

**Combined** when this turn (or the live map session) clearly has map UI / UI terminal / map PTY **and** keep-draining / catch-up / `work next --wait` together. Or if user explicitly asks for combined mode. Then read **both** `drain.md` and `learn.md`.

If drain and learn appear without a combined signal → ask one line which branch, or take the first/stronger signal. Then read only that file.

## 3. Branch

- Combined → read **`drain.md`** and **`learn.md`**
- Learn → read **`learn.md`**
- Drain (default) → read **`drain.md`**

### Combined session rules

- Background: one `snowshoe work next --json --wait` (optional `--wait-timeout 0`). Foreground: learn/map conversation in the same session / map PTY / UI terminal — do **not** kill the background waiter to study the map.
- Align with `drain.md` interactive `--wait`: at most **one** waiter; gates/work return immediately; when the waiter returns work, handle it then restart the waiter; never start a second waiter.

## 4. Optional feedback

`snowshoe feedback add --json` (stdin `{ "text": "...", "command": "optional" }`) appends a local note about Snowshoe. Use only if the tool is unclear, frustrating, or you have a concrete idea. Not required. Do not log secrets or lease tokens. Then continue drain, learn, or combined.
05

Trust audit

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

LayerWhat it checksResult
L0Provenance & inventoryWARN
L1Static analysis of the codeNA
L2Instruction surface (what it tells the agent)PASS
L3Class-specific surfacePASS
L4Behavioural (sandbox)SKIPPED

What the source does

Filesystem
none-observed
Network
none-observed
Shell
none-observed
Dependencies
pinned
Secrets in source
none-found

Findings (2)

MEDIUMInventory / provenance · inv.symlink · CWE-1104
.cursor/skills/snowshoe
.cursor/skills/snowshoe
Why it matters. link not followed
INFOPrompt injection · prompt.persistence · CWE-94, CWE-1427
install.md:35
# ensure ~/.bun/bin is on PATH (bun link alone does not edit your shell PATH)
Why it matters. instructs the agent to persist itself in the user's environment

Gates applied: no_behavioural_pass.

Audited 2026-10-10 · audit v0.4.1 · source sha c1fbd5a3e704full audit observations/trust-audit/skill/igorkravcenko__snowshoe.json · Report an issue / request a re-scan
06

Audit history

Every audit this skill has had.

DateSourceVerdictGradeScoreChange
2026-10-10c1fbd5a3e704SAFEB89first audit
07

Questions

What does the snowshoe skill do?

Don't let your agents outrun your understanding. Keep your footing. A git-native personal comprehension map of a repository — tracks what goes stale after pull. Catch up after pull.

Is snowshoe safe to install?

The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.

What can snowshoe access on my machine?

The audit observed no filesystem, network or shell use at all in its source.

What do I need installed to use snowshoe?

Its own instructions reference snowshoe. Dependencies are pinned to exact versions.

Which assistants does snowshoe work with?

Its documentation mentions cursor. That is what the text claims, not a compatibility test we ran.

How current is this page?

The grade is for one exact copy of the source (c1fbd5a3e704), read on 2026-10-10. The repository is watched, and a new audit runs when it changes — this is the first audit.

Advertisement