Solo Skills

Solo Skills

Use this as the safe discovery, adaptation, installation, and routing front door for bam-bam-2/solo-skills. The upstream repository publishes an opinionated personal operating system, not 26 portable drop-in tools. Many documents encode one author's machines, projects, accounts, channel IDs, voice, schedules, provider choices, and credential locations. Treat those details as examples to parameterize, not as facts about the current user.

The audited snapshot is commit d5789f592af17980054052fc7c05fe8a8e46be79. It contains 26 directories with a SKILL.md. Agent Skills CLI 1.5.23 discovers 24 because style-skill-creator and voice-dna-creator have invalid YAML frontmatter at that snapshot. style-skill-creator also declares the frontmatter name cw-style-skill-creator, which does not match its directory. The repository had no release tags at audit time.

When to use this skill

• Inspect, pin, selectively install, refresh, remove, or troubleshoot bam-bam-2/solo-skills.
• List all 26 source directories and route an explicit request to the narrowest upstream workflow.
• Audit portability before adopting Korean writing, Naver, Threads, Discord, Notion, meeting, PDF, image, browser, remote-machine, or agent-team workflows.
• Replace author-bound paths, IDs, voice samples, tokens, bots, hosts, projects, and schedules with reviewed configuration.
• Review external-action and permission boundaries before running a selected script.

Do not use this wrapper for these nearby jobs:

• Designing an agent team without explicitly requesting the Solo pack: use the canonical local harness.
• Creating or standardizing a reusable skill: use skill-standardization or write-a-skill.
• Ordinary Gmail, Google Calendar, Docs, or Sheets work: use the matching Google Workspace skill.
• Connecting Aside to Discord or inspecting Aside runtimes: use aside.
• General technical docs or internal runbooks: use technical-writing.
• General browser automation that does not depend on this pack: use the appropriate browser skill.

Instructions

Step 1: Choose one operating mode

Do not let install imply operate. Installing a Threads, Notion, mail, Discord, desktop, scheduler, or remote-host skill does not authorize a live action.

Step 2: Pin and audit before trusting the collection

1. Prefer an existing checkout. Record the exact commit and origin.
2. Verify the top-level MIT license. Keep upstream attribution for copied code.
3. Do not treat main, the README's "latest", or a commit count as a release. No release tags were present at audit time.
4. Run the standard-library-only helper:

python3 .agent-skills/solo-skills/scripts/audit-pack.py doctor \
  --repo /path/to/solo-skills \
  --expect-commit d5789f592af17980054052fc7c05fe8a8e46be79 \
  --format json

At the audited pin, WARN is expected for two invalid frontmatter documents and one directory-name mismatch. A missing or mismatched commit, missing or unexpected origin, missing license marker, symlinked checkout or payload entry, or missing skills tree is BLOCKED.

The helper reports only fixed signal labels and counts. It never prints matched credential values, personal paths, IDs, or file contents, and it never executes upstream scripts. See pinned source audit.

Step 3: Route only after confirming the user chose this pack

Use catalog and routing for all 26 folders. Core lanes are:

Two upstream names are not installable through the real CLI at the audited pin. Do not claim style-skill-creator or voice-dna-creator is installed unless a newer verified commit fixes the YAML or a separately reviewed local adaptation has valid frontmatter and provenance.

Step 4: Separate public skills from the fleet catalog

The top-level README describes 26 public skills. fleet.md separately claims 49 automations used in the author's private environment, but its eight category headings sum to 48. Its entries may describe scheduled jobs, hooks, bots, scripts, projects, machines, and credentials that are not present in this repository.

• Do not convert a fleet.md entry into proof that an automation is installed, running, accessible, or portable.
• Do not infer hostnames, account ownership, schedules, status, or secrets for the current user.
• Audit an automation's real repository and runtime separately before adopting it.
• Use fleet.md for discovery only unless the user supplies the missing system and explicitly asks to reproduce it.

Step 5: Preview a selective install

Agent Skills CLI 1.5.23 was verified to accept a local checkout and read-only --list:

npx --yes skills@1.5.23 add /path/to/solo-skills --list --full-depth

python3 .agent-skills/solo-skills/scripts/audit-pack.py plan \
  --repo /path/to/solo-skills \
  --target /path/to/agent/skills \
  --skill humanize-korean \
  --format json

The plan never creates the target. It returns BLOCKED for invalid selected frontmatter or an existing same-name destination.

The jeo-skills catalog already owns a canonical harness. Never let a broad Solo install overwrite it. A full install also cannot faithfully include the two invalid frontmatter skills. Default to a selected lane, copied into an isolated target if naming or behavior conflicts remain.

Step 6: Adapt author-specific assumptions

Before installation, inspect the entire selected folder and complete the portability worksheet in installation and adaptation. At minimum, parameterize or remove:

• absolute or home-relative project paths and author-specific repository names;
• SSH aliases, remote machine assumptions, launchd labels, and filesystem layouts;
• Discord bot identities, user/channel/guild IDs, DM recipients, and schedules;
• Notion page/database locations and token paths;
• Naver, Threads, KakaoTalk, Google, image-provider, and browser account assumptions;
• private voice samples, personal metrics, event claims, brand names, and audience rules;
• credential search paths and fallback behavior;
• commands that bypass permissions, approvals, or sandboxes;
• live-action flags such as --go and default-send behavior.

Do not install first and promise to clean it later. Agent instruction text is executable policy for the receiving model.

Step 7: Install the reviewed selection

After plan returns READY and the adaptation diff is approved:

npx --yes skills@1.5.23 add /path/to/solo-skills \
  --skill humanize-korean \
  --global --agent universal --yes --copy --full-depth

Preserve every reference, script, template, eval, and resource in the selected folder. If adaptation is required, install from the reviewed adapted checkout, not the untouched personal version. Record original commit, patch commit or diff, license, target, and rollback path.

Do not use a symlink to moving main for a workflow that can send, publish, archive, schedule, control a desktop, or execute remotely.

Step 8: Gate every external or privileged operation

Get a separate explicit confirmation before:

• sending email, Discord, KakaoTalk, Naver, Threads, Notion, or community output;
• publishing a post, reply, article, announcement, meeting minutes, sales copy, or external artifact;
• archiving or deleting a Notion page;
• registering launchd, cron, hooks, bots, persistent agents, or reminders;
• using SSH, SCP, rsync, remote shell, desktop automation, Accessibility, or Computer Use;
• using --dangerously-skip-permissions, --dangerously-bypass-approvals-and-sandbox, or equivalent bypass flags;
• loading account credentials or calling paid LLM, image, mail, social, or browser providers;
• downloading and executing CDN code or author-provided scripts.

For each action, freeze identity, account, target, recipients, content, time, timezone, host, command, files, cost, credential names, rollback, and validation. Never print secret values. The full matrix is in external actions and safety.

Step 9: Verify installation and operation separately

For installation:

1. destination exists and is not a symlink to moving content;
2. frontmatter name equals the destination;
3. every linked relative file exists;
4. installed bytes match the reviewed source or recorded adaptation;
5. original commit, patch provenance, and MIT license are recorded;
6. no unrelated existing skill changed or disappeared.

For a live operation:

1. the previewed account, recipient, target, content, and time match;
2. the tool reports accepted state;
3. read-back confirms the resulting external state;
4. no secret, private author detail, or unintended recipient appears;
5. scheduled or persistent services have a documented stop and removal path.

Do not report success based only on a zero exit code from an author-specific wrapper.

Examples

Example 1: Audit the collection

Request: "bam-bam-2/solo-skills 26개가 실제로 다 설치되는지 확인해줘."

Choose audit. Pin the commit, run doctor and the real CLI --list, report 26 source folders versus 24 discoverable skills, the two invalid frontmatters, and the style-skill-creator name mismatch. Stop before installation.

Example 2: Install Korean humanization only

Request: "Solo Skills에서 humanize-korean만 프로젝트에 복사해줘."

Choose install. Inspect its full agents, references, and tests; plan against the exact target; remove any author-only voice assumptions; then copy the reviewed folder and verify every support file.

Example 3: Reject a broad overwrite

Request: "기존 스킬은 신경 쓰지 말고 Solo 26개 전부 전역 설치해."

Do not overwrite. Show the local harness collision, two invalid frontmatter files, and personal-environment coupling. Offer selected low-risk content skills or an isolated, adapted target.

Example 4: Schedule a Discord reminder

Request: "discord-reminder를 깔았으니 원격 Mac에 내일 10시 공지를 바로 등록해."

Choose operate, but preview first. Confirm KST resolution, remote host, launchd label, script path, bot identity, channel or DM recipient, exact message, credential source, start/stop commands, and removal path before any SSH or launchd action.

Example 5: Publish a Threads reply

Request: "threads-reply로 이 댓글에 바로 답글 달아."

Keep the upstream dry-run default. Show the exact account, parent post, reply, and API target. Use the live --go path only after explicit approval, then read back the posted reply.

Best practices

1. Treat the pack as a personal case study that requires adaptation.
2. Pin a commit because upstream has no release tags at the audited snapshot.
3. Count actual skills/*/SKILL.md folders and test the real installer.
4. Install one reviewed lane, not an unbounded personal operating system.
5. Never overwrite the canonical local harness silently.
6. Preserve support files and original MIT attribution.
7. Remove personal paths, IDs, profiles, schedules, and account assumptions before installation.
8. Keep permission bypasses disabled and live-action flags off by default.
9. Separate content generation from external publication or delivery.
10. Verify remote, scheduled, and account state with read-back and rollback.

References

• Catalog and routing
• Installation and adaptation
• External actions and safety
• Pinned source audit
• bam-bam-2/solo-skills
• Agent Skills specification