复制安装命令
用 Codex 或 Claude 安装复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它先审查 Skill 页面再帮你安装。
复制前请先查看来源、License 和安全提示。
23 skills for AI coding agents — persistent memory, session awareness, and collaborative tools.
用 Codex 或 Claude 安装复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它先审查 Skill 页面再帮你安装。
复制前请先查看来源、License 和安全提示。
来源文件:README.md
23 skills for AI coding agents — persistent memory, session awareness, and collaborative tools.
In Claude Code — no bun, no git, nothing installed first:
/plugin marketplace add Soul-Brews-Studio/arra-oracle-skills-cli
/plugin install oracle-skills@oracle-skills
In a terminal — for any of the 19 supported agents:
bunx --bun github:Soul-Brews-Studio/arra-oracle-skills-cli#alpha install -g -y -p full
-p full gives you every stable skill — see the Profiles table below for the current count. Everything else is a flag on that same command.
| want | add |
|---|---|
| a smaller set | -p standard (20) · -p minimal (7) |
| experimental too | -p lab — same as full whenever nothing is currently flagged experimental |
| a few extra skills | -s recap rrr trace — added on top of the profile |
| a specific agent | --agent codex · cursor · opencode · gemini-cli · claude-code |
| several at once | --agent claude-code codex opencode |
| slash-command stubs | --with-commands — Codex, OpenCode and Gemini need these |
| this project only | drop -g → installs to ./.claude/skills/ |
| federated agents | --with-thclaws / -a thclaws / --all-detected — never auto-detected (#330) |
Update the way you installed: plugin → /plugin update oracle-skills@oracle-skills; terminal → re-run the command. Don't update through an installed arra-oracle-skills binary — it carries its own frozen copy of the skills, so once it's older than yours, "update" writes the old set back. Always fetch from GitHub.
npm is a lagging mirror (publishing is manual). npx arra-oracle-skills@latest works; never pin an exact -alpha version from git history — not every bump is published.
zsh note: -s needs each name as its own word, and zsh doesn't split $VARS — write names literally or use ${=NAMES}, or you'll silently get only the profile.
19 agents: Claude Code, Codex, OpenCode, Cursor, Gemini CLI, Amp, Kilo Code, Roo Code, Goose, Antigravity, GitHub Copilot, OpenClaw, Droid, Windsurf, Cline, Aider, Continue, Zed, thClaws
📚 23 skills
| # | Skill | Type | Description |
|---|---|---|---|
| 1 | about-oracle | skill + subagent | What is Oracle |
| 2 | awaken | skill | Guided Oracle birth and awakening ritual |
| 3 | bampenpien | skill | บำเพ็ญเพียร |
| 4 | bud | skill | Create a new oracle via maw bud |
| 5 | create-shortcut | skill | Create local skills as shortcuts |
| 6 | dig | skill | Mine Claude Code sessions |
| 7 | feel | skill | Capture how the system feels |
| 8 | forward | skill | Hand off the current session to the next one |
| 9 | go | skill | Manage Oracle skills |
| 10 | incubate | skill | Clone or create repos for active development |
| 11 | learn | skill + subagent | Explore a codebase with parallel Haiku… |
| 12 | oracle-cheatsheet | skill | Generate a copy-paste cheat sheet from the… |
| 13 | oracle-family-scan | skill + code | Oracle Family Registry |
| 14 | oracle-prism | skill | Multi-perspective analysis |
| 15 | philosophy | skill | Display Oracle philosophy |
| 16 | project | skill + code | Clone and track external repos |
| 17 | psi | skill | Attach a code repo's ψ vault to a caretaker… |
| 18 | recap | skill + code | Session orientation and awareness |
| 19 | resonance | skill | Capture a resonance moment |
| 20 | rrr | skill | Create a session retrospective with an AI… |
| 21 | trace | skill | Find projects, code, and knowledge across… |
| 22 | where-we-are | skill | Session awareness |
| 23 | who-are-you | skill | Know ourselves |
| Profile | Count | Skills |
|---|---|---|
| minimal | 7 | about-oracle, forward, go, recap, rrr, trace, who-are-you |
| standard | 22 | about-oracle, awaken, bampenpien, bud, create-shortcut, dig, feel, forward, go, incubate, learn, oracle-cheatsheet, oracle-family-scan, oracle-prism, oracle-write-complete-book, psi, recap, resonance, rrr, trace, where-we-are, who-are-you |
| full | 23 | all |
| lab | 23 | all |
Switch anytime: /go standard, /go full, /go lab
install [options] # install skills (default profile: minimal)
uninstall [options] # remove installed skills
select [options] # interactive skill picker
list [options] # show installed skills
profiles [name] # list profiles
agents # list supported agents
about # version + status
Archived skills — excluded from every profile — live in their own repository:
→ https://github.com/Soul-Brews-Studio/arra-oracle-skills-archive
Moved out of this repo starting 2026-08-22. Nothing was deleted: their full history
remains in this git log, and src/skills/.archive/MOVED.md is the breadcrumb.
git clone https://github.com/Soul-Brews-Studio/arra-oracle-skills-archive
cp -R arra-oracle-skills-archive/skills/<name> ~/.claude/skills/
arra install -s <zombie-name> no longer resolves — those skills are not bundled
in the CLI any more. That path was removed deliberately, not by accident.
Nat Weerawan — Soul Brews Studio · MIT
name: go
description: Manage Oracle skills — list, install, remove, find, switch profiles, update. Use when user says "go", "install skill", "remove skill", "find skill", "switch profile", "go list", "go update", "go install", "go find". Single entry point for all skill management.
argument-hint: "[list] | <standard|full|lab|cleanup|update> | install|remove|find <skill...>"
disable-model-invocation: trueSwitch gear. Single source of truth.
/go # all skills — profile tier + installed status
/go list # same as /go (no args)
/go minimal # newcomer essentials (7 skills, default)
/go standard # daily driver (13 skills)
/go full # all stable (excludes lab-only experiments)
/go lab # everything including experimental
/go cleanup # remove ALL skills → fetch latest → fresh install
/go update # check for new version + upgrade in-place
/go install team-agents # cherry-pick install specific skills
/go remove watch # uninstall specific skills
/go find feel # search all available skills by name
⚠ NEW (#285):
/go <profile>now ALIGNS your installed skills to the target profile. arra-managed skills NOT in the target profile will be removed. External / non-arra skills are never touched. A pre-removal diff is shown before any deletion occurs.
Before running any command, detect the CLI path. It may not be in $PATH on all machines.
Two variables, because reading and writing have different safety requirements.
# $ARRA — READ-ONLY commands (profiles, list, --version). The local binary is fine
# here: it is fast, and a slightly stale profile listing is harmless.
if command -v arra-oracle-skills &>/dev/null; then
ARRA="arra-oracle-skills"
elif [ -x "$HOME/.bun/bin/arra-oracle-skills" ]; then
ARRA="$HOME/.bun/bin/arra-oracle-skills"
else
ARRA="$HOME/.bun/bin/bunx --bun github:Soul-Brews-Studio/arra-oracle-skills-cli#alpha"
fi
# $ARRA_INSTALL — anything that WRITES skills to disk. Always from GitHub.
ARRA_INSTALL="$HOME/.bun/bin/bunx --bun github:Soul-Brews-Studio/arra-oracle-skills-cli#alpha"
Why they differ. The installed binary carries its own frozen copy of every skill — the
set as of the day you installed it. Installing through it therefore writes those skills
over whatever you have now. Once the binary is older than your skills (normal after a few
weeks) every "install" is really a downgrade; a 26.5.15 binary was seen overwriting
26.7.25 skills. Reading through it costs you nothing, so keep the fast path for reads and
take the network hit only when writing.
Use $ARRA for reads and $ARRA_INSTALL for installs/updates below.
Parse the user's /go arguments and run the matching $ARRA command.
/go or /go list — show all skills with profile + installed status + typeWrite this script to /tmp/go-list.sh and run it with bash /tmp/go-list.sh. The script auto-discovers profile membership from the CLI — zero hardcoded skill names.
#!/bin/bash
SKILLS_DIR="$HOME/.claude/skills"
ARRA="$HOME/.bun/bin/arra-oracle-skills"
VERSION=$($ARRA --version 2>/dev/null || echo "?")
# Auto-get skill lists from CLI (no hardcoding)
get_profile_skills() {
$ARRA profiles "$1" 2>/dev/null | grep "Skills:" | sed 's/.*Skills: //' | tr ',' '\n' | sed 's/^ *//' | tr '\n' ' '
}
STANDARD=$(get_profile_skills standard)
FULL=$(get_profile_skills full)
LAB_RAW=$(get_profile_skills lab)
# Lab-only = in lab but not in full
LAB=""
for s in $LAB_RAW; do
echo " $FULL " | grep -q " $s " || LAB="$LAB $s"
done
# Minimal-only = in minimal but not in standard
MINIMAL=$(get_profile_skills minimal)
MINIMAL_ONLY=""
for s in $MINIMAL; do
echo " $STANDARD " | grep -q " $s " || MINIMAL_ONLY="$MINIMAL_ONLY $s"
done
ALL_ARRA="$STANDARD $LAB $MINIMAL_ONLY"
detect_type() {
local dir="$SKILLS_DIR/$1"
[ -d "$dir" ] || return
local has_code=false has_agent=false
[ -d "$dir/scripts" ] && has_code=true
grep -qlE 'Agent\(|subagent|TeamCreate|SendMessage' "$dir/SKILL.md" 2>/dev/null && has_agent=true
if $has_code && $has_agent; then echo "[code+agent]"
elif $has_code; then echo "[code]"
elif $has_agent; then echo "[agent]"
fi
}
get_desc() {
local f="$SKILLS_DIR/$1/SKILL.md"
[ -f "$f" ] || return
awk '/^---$/{n++; next} n==1{print} n>=2{exit}' "$f" \
| grep -E '^description:' | head -1 \
| sed "s/description: *//; s/['\"]//g; s/\[core\].*G-SKLL | //" \
| cut -c1-40
}
echo "Oracle Skills v$VERSION"
echo ""
NUM=0; INSTALLED=0
std_count=$(echo $STANDARD | wc -w | tr -d ' ')
lab_count=$(echo $LAB | wc -w | tr -d ' ')
min_count=$(echo $MINIMAL_ONLY | wc -w | tr -d ' ')
for tier in STANDARD LAB MINIMAL_ONLY; do
case $tier in
STANDARD) skills="$STANDARD"; label="standard"; count=$std_count ;;
LAB) skills="$LAB"; label="lab"; count=$lab_count ;;
MINIMAL_ONLY) skills="$MINIMAL_ONLY"; label="minimal-only"; count=$min_count ;;
esac
[ -z "$(echo $skills | tr -d ' ')" ] && continue
printf " ─── %s (%s) " "$label" "$count"
printf '─%.0s' $(seq 1 $((52 - ${#label}))); echo ""
for name in $skills; do
NUM=$((NUM+1))
if [ -d "$SKILLS_DIR/$name" ]; then
mark="✓"; INSTALLED=$((INSTALLED+1))
else
mark="✗"
fi
tag=$(detect_type "$name")
desc=$(get_desc "$name")
if [ -n "$desc" ]; then
printf " %2d %s %-24s %-14s %s\n" "$NUM" "$mark" "$name" "$tag" "$desc"
else
printf " %2d %s %-24s %s\n" "$NUM" "$mark" "$name" "$tag"
fi
done
echo ""
done
# External (non-arra) skills
EXT=0; EXT_NAMES=""
for dir in "$SKILLS_DIR"/*/; do
[ -d "$dir" ] || continue
name=$(basename "$dir"); [ "$name" = ".trash" ] && continue
is_arra=false
for a in $ALL_ARRA; do [ "$name" = "$a" ] && is_arra=true && break; done
[ "$is_arra" = "false" ] && { EXT=$((EXT+1)); EXT_NAMES="$EXT_NAMES $name"; }
done
if [ "$EXT" -gt 0 ]; then
printf " ─── external (%d) " "$EXT"
printf '─%.0s' $(seq 1 44); echo ""
for name in $EXT_NAMES; do
NUM=$((NUM+1)); INSTALLED=$((INSTALLED+1))
tag=$(detect_type "$name")
desc=$(get_desc "$name")
if [ -n "$desc" ]; then
printf " %2d ✓ %-24s %-14s %s\n" "$NUM" "$name" "$tag" "$desc"
else
printf " %2d ✓ %-24s %s\n" "$NUM" "$name" "$tag"
fi
done
echo ""
fi
echo " $INSTALLED/$NUM installed · zombies hidden (--zombies to show)"
echo " Cherry-pick: arra-oracle-skills install -g -s <name> -y"
echo " Switch: /go standard | /go lab"
Auto-detected columns:
arra-oracle-skills profiles <name> CLI output (zero hardcoded names)[code] (scripts/), [agent] (Agent patterns), [code+agent] (both)$HOME/.claude/skills/<name>/ exists--zombies flag: append zombie skills section. Only shown when explicitly requested.
/go <profile> — switch profile$ARRA_INSTALL install -g --profile <name> -y
Profiles: minimal, standard, full, lab
/go minimal → $ARRA_INSTALL install -g --profile minimal -y/go standard → $ARRA_INSTALL install -g --profile standard -y/go full → $ARRA_INSTALL install -g --profile full -y/go lab → $ARRA_INSTALL install -g --profile lab -yNote: passing
--profileexplicitly triggers alignment — arra-managed skills NOT in the target are removed automatically. External skills are never touched.
/go cleanup — fresh install (safe)Crosscheck installed skills, remove stale arra-managed ones, fetch latest, reinstall. External skills are never touched.
Step 1: Crosscheck — list all installed skills, classify by installer: field in SKILL.md:
SKILLS_DIR="$HOME/.claude/skills"
LATEST=$(curl -s https://api.github.com/repos/Soul-Brews-Studio/arra-oracle-skills-cli/tags | grep -m1 '"name"' | cut -d'"' -f4)
echo "📋 Crosscheck (latest: $LATEST):"
ARRA_COUNT=0; EXT_COUNT=0; STALE_COUNT=0; CONFLICT_COUNT=0
for dir in "$SKILLS_DIR"/*/; do
[ -d "$dir" ] || continue
name=$(basename "$dir")
version=$(grep -o 'v[0-9][0-9.]*' "$dir/SKILL.md" 2>/dev/null | head -1)
installer=$(grep 'installer:' "$dir/SKILL.md" 2>/dev/null | head -1)
if echo "$installer" | grep -q "arra-oracle"; then
# Arra-managed skill — check version
if [ "$version" = "$LATEST" ]; then
echo " ✓ arra: $name ($version)"
ARRA_COUNT=$((ARRA_COUNT + 1))
else
echo " ⚠️ stale: $name ($version → $LATEST)"
STALE_COUNT=$((STALE_COUNT + 1))
fi
else
echo " ○ external: $name — will keep"
EXT_COUNT=$((EXT_COUNT + 1))
fi
done
echo ""
echo " Summary: $ARRA_COUNT ok, $STALE_COUNT stale, $EXT_COUNT external"
Step 2: Combined table — crosscheck + usage in ONE table. Mine session JSONL files, then display everything together:
# Mine usage data from all sessions
TOTAL=0
for jsonl in ~/.claude/projects/*/*.jsonl; do
[ -f "$jsonl" ] || continue
TOTAL=$((TOTAL + 1))
done
Build the combined table. For each of the 29 arra skills, show: profile tier, installed status, version, status, and usage count from session mining.
📋 Skills Overview (29 arra + N external) — $TOTAL sessions mined:
# Skill Profile Installed Version Status Usage
── ──────────────────────── ────────── ────────── ───────── ──────────── ─────
1 about-oracle standard ✓ v3.7.2 ✓ ok 2
2 awaken standard ✓ v3.7.2 ✓ ok 7
4 contacts lab ✓ v3.7.2 ✓ ok 5
5 create-shortcut lab ✗ — — 3
6 dig standard ✓ v3.7.2 ✓ ok 6
7 dream lab ✗ — — 5
8 feel lab ✗ — — 4
9 forward standard ✓ v3.7.2 ✓ ok 4
10 go standard ✓ v3.7.2 ✓ ok 3
11 inbox lab ✓ v3.7.2 ✓ ok 4
12 incubate full ✓ v3.7.2 ✓ ok 6
13 learn standard ✓ v3.7.2 ✓ ok 7
14 oracle-family-scan standard ✓ v3.7.2 ✓ ok 4
15 oracle-soul-sync-update standard ✓ v3.7.2 ✓ ok 3
16 philosophy full ✗ — — 4
17 project full ✗ — — 6
18 recap standard ✓ v3.7.2 ✓ ok 7
19 resonance full ✗ — — 6
20 rrr standard ✓ v3.7.2 ✓ ok 7
21 schedule lab ✗ — — 3
22 standup standard ✓ v3.7.2 ✓ ok 3
23 talk-to standard ✓ v3.7.2 ✓ ok 6
24 team-agents lab ✗ — — 1
25 trace standard ✓ v3.7.2 ✓ ok 5
26 vault lab ✗ — — 5
27 where-we-are full ✗ — — 4
28 who-are-you full ✓ v1.0.22 ⚠️ stale 6
29 xray standard ✓ v3.7.2 ✓ ok 4
External (will keep):
○ drink, mawjs, mawjs-local, ultrathink
💡 Skills with 0 usage might not need to be in your profile.
How to get usage counts: for each skill, count sessions containing /$skill:
for skill in about-oracle awaken contacts create-shortcut \
dig dream feel forward go inbox incubate learn oracle-family-scan \
oracle-soul-sync-update philosophy project recap resonance rrr \
schedule standup talk-to team-agents trace vault where-we-are who-are-you xray; do
count=$(grep -rl "/$skill" ~/.claude/projects/*/*.jsonl 2>/dev/null | wc -l)
echo "$count $skill"
done | sort -rn
Status legend:
✓ ok — arra-managed, current version⚠️ stale — arra-managed but outdated (needs update)— — not installed (available in higher profile)Step 3: Confirm — now with full context:
Proceed with cleanup?
- Conflicts will be replaced (backed up to .bak)
- External skills kept untouched
- Which profile? [standard / full / lab]
Step 4: Clean + reinstall (only after user confirms):
# Uninstall arra-managed via CLI
$ARRA uninstall -g -y
# For each conflict skill: rename to .bak (Nothing is Deleted)
for name in [conflicting skills]; do
mv "$SKILLS_DIR/$name" "$SKILLS_DIR/${name}.bak.$(date +%s)"
done
# Fresh install at latest
LATEST=$(curl -s https://api.github.com/repos/Soul-Brews-Studio/arra-oracle-skills-cli/tags | grep -m1 '"name"' | cut -d'"' -f4)
~/.bun/bin/bunx --bun arra-oracle-skills@github:Soul-Brews-Studio/arra-oracle-skills-cli#$LATEST install -g -y
Output:
🧹 Cleanup complete!
Kept: [N] external skills
Replaced: [N] conflicts (backed up to .bak)
Installed: [N] fresh at $LATEST
Restart required.
When to use:
[hidden] flags persisting after unhide/go update — pull the newest skills from GitHubNever update through the installed $ARRA binary. It ships its own copy of the skills,
frozen at whatever version you installed it — so $ARRA install re-writes those frozen
skills over your current ones. When the binary is older than your skills (the normal case
after a few weeks), "update" quietly becomes a downgrade. Observed live: a
26.5.15 binary overwriting 26.7.25 skills.
Always fetch from GitHub instead, so the newest skills come from the branch rather than from whatever is cached on disk:
CURRENT=$($ARRA --version 2>/dev/null || echo "unknown")
LATEST=$(curl -s https://api.github.com/repos/Soul-Brews-Studio/arra-oracle-skills-cli/tags | grep -m1 '"name"' | cut -d'"' -f4)
echo " CLI binary: $CURRENT"
echo " Latest tag: $LATEST"
# The update itself — always from GitHub, never from $ARRA
$ARRA_INSTALL install -g -y
#alpha is the branch every cut lands on first; it is also the repo's default branch, so
this is the same code the plugin marketplace serves. Pin a tag (#$LATEST) instead when
you deliberately want a specific release.
Already on the plugin? If claude plugin list shows oracle-skills@oracle-skills, that
install updates itself — run /plugin update oracle-skills@oracle-skills and stop here.
Running both leaves two copies of every skill and it stops being obvious which one loads.
Refreshing a specific set — -s takes a LIST of names, so each name must arrive as its
own argument. In zsh (the default shell here) $VAR does not word-split, so a variable
holding names collapses into one bogus argument and the CLI silently installs only the
default profile:
NAMES="dig learn talk-to"
… install -g -y -s $NAMES # zsh: ONE argument → wrong, installs default profile only
… install -g -y -s ${=NAMES} # zsh: three arguments → correct
… install -g -y -s dig learn talk-to # always correct, in any shell
Also note -s is additive on top of the default profile — it adds the named skills, it
does not restrict the install to them.
Skills are only half of what goes stale. The arra-oracle MCP server is wired
separately in ~/.claude.json, so it drifts on its own — a real case had it
pinned 161 commits / 30 days behind while /go update still reported
everything synced, because the skills matched. Check both, or "synced" is a
half-truth.
python3 - <<'PY'
import json, os, re, subprocess
cfg = (json.load(open(os.path.expanduser('~/.claude.json'))).get('mcpServers') or {}).get('arra-oracle')
if not cfg:
print(' arra-oracle MCP: not configured — nothing to check'); raise SystemExit
args = ' '.join(cfg.get('args') or [])
# 1. Local checkout — drift is commits behind upstream.
repo = next((a for a in (cfg.get('args') or []) if a.startswith('/')), None)
if repo:
while repo != '/' and not os.path.isdir(os.path.join(repo, '.git')):
repo = os.path.dirname(repo)
if repo != '/':
subprocess.run(['git', '-C', repo, 'fetch', '--quiet', 'origin'], check=False)
behind = subprocess.run(['git', '-C', repo, 'rev-list', '--count', 'HEAD..@{u}'],
capture_output=True, text=True).stdout.strip() or '?'
print(f' arra-oracle MCP: local checkout {repo}')
print(f' {"up to date" if behind == "0" else behind + " commits behind upstream — git -C " + repo + " pull"}')
raise SystemExit
# 2. Git-pinned bunx — drift is the pinned ref vs upstream HEAD.
m = re.search(r'github:([\w.-]+/[\w.-]+)#([\w.-]+)', args)
if m:
slug, ref = m.groups()
head = subprocess.run(['git', 'ls-remote', f'https://github.com/{slug}', 'HEAD'],
capture_output=True, text=True).stdout.split()
print(f' arra-oracle MCP: pinned {slug}#{ref}')
print(f' upstream HEAD {head[0][:12] if head else "unknown"} — repin if this is stale')
raise SystemExit
# 3. Unpinned bunx — resolves from bun's cache, so version is whatever was cached.
print(' arra-oracle MCP: unpinned bunx — resolves from cache, version unknowable')
print(' pin a tag, or clear with: bun pm cache rm')
PY
Report drift alongside the skills version; never auto-edit ~/.claude.json —
that file holds credentials for every MCP server the user has, and a wrong write
costs more than a stale server. Print the fix and let them run it.
/go update → fetch latest from GitHub and reinstall in place-p <name> only if you also want to switch tiers/go install <skill...> — cherry-pick specific skills$ARRA_INSTALL install -g -s <skill...> -y
/go install team-agents → $ARRA_INSTALL install -g -s team-agents -y/go install dig hey xray → $ARRA_INSTALL install -g -s dig hey xray -yWorks for ANY skill — standard, lab, or even zombie. Does not change your profile; just adds the named skills on top of whatever you have.
/go remove <skill...> — uninstall specific skills$ARRA uninstall -g -s <skill...> -y
/go remove watch → $ARRA uninstall -g -s watch -y/go find <query> — search available skills$ARRA profiles lab 2>/dev/null
Show all skills from the lab profile (the superset) and highlight any matching the query. If the user says /go find feel, search the profile output for "feel" and show where it lives (which tier, installed or not).
/go find feel → show feel is in lab tier, not installed, description/go find team → show team-agents in standard tier, installed| Profile | Count | Description |
|---|---|---|
| minimal | 7 | Newcomer essentials — lite lifecycle, trace, update (token-optimized) |
| standard | 13 | Daily driver — essential Oracle skills |
| full | ~30 | All stable skills (excludes lab-only + minimal-only lite variants) |
| lab | ~48 | Everything including experimental (excludes minimal-only lite variants) |
-g — global (user-level) skills-y — skip confirmationgo is always preserved — it's in every profileARGUMENTS: $ARGUMENTS
评论 (0)
暂无评论,成为第一个评论者吧!