Dobeu-tech-eco

find-skills

Rescan the repo, audit every improvement option (skills, database tooling, hooks, MCP servers, plugins, commands, subagents, permissions, CI, docs), ask one to three questions, recommend a ranked set, and wire in the picks on a yes. Use when the user says find skills or how can I improve this setup.

Dobeu-tech-eco 0 Updated 1w ago
GitHub

Install

npx skillscat add dobeu-tech-eco/dts2-marketplace/find-skills

Install via the SkillsCat registry.

SKILL.md

Directory edition: do not run bundled scripts or hooks, a package launcher, or a download piped into a shell. Do not read tokens or credential files from the machine. Carry out the step with Read, Edit, Grep, Glob, and git or gh, and write the files the step names.

Find skills — rescan, audit, recommend, wire in

Runnable at any point in a project, not only during the setup deep interview. It never changes anything without a yes.

1. Rescan

Characterize the repo with the existing read-only scripts:

  • python3 "${CLAUDE_PLUGIN_ROOT}/scripts/scope-check.py" . (is there a harness, a workspace, an end state)
  • python3 "${CLAUDE_PLUGIN_ROOT}/scripts/repo-size.py" . (size and tier)
  • python3 "${CLAUDE_PLUGIN_ROOT}/scripts/env-detect.py" (tools available on this machine)
  • python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/improve-scan.py" --cwd . (read-only audit: facts, an inventory of hooks, MCP servers, plugins, commands, agents and permissions, and recommendations)

The facts come from improve-scan.py: is_brownfield, is_large, has_tests, has_frontend, has_pr (a GitHub remote or .github/), has_ci, has_marketing (a landing or pricing page), has_agent_docs (CLAUDE.md, AGENTS.md), and the database facts has_database, has_prisma, has_supabase, has_drizzle, has_postgres, has_sqlite. Leave a fact unset when unknown.

2. Ask

Ask one to three clarifying questions with AskUserQuestion about what the user wants next (for example: ship a feature, fix bugs, polish the UI, launch). Skip a question the scan already answers.

3. Recommend

Run python3 "${CLAUDE_PLUGIN_ROOT}/scripts/skill-suggest.py" --facts '<the facts from improve-scan as JSON>'. It returns a ranked list of slugs from the mattpocock companion set (grill-me, improve-codebase-architecture, writing-for-agents, pr, ask-matt, research, diagnosing-bugs), resolve-merge-conflicts from warpdotdev/common-skills, plus design, test and marketing skills. Add any already-installed skills that fit the user's answers.

Merge those skills with the improve-scan recommendations. Database tooling is a featured category: show its items near the top, right after priority 1. Show at most twelve, grouped by category (skills, database, hooks, mcp, plugins, commands, subagents, permissions, ci, docs), each with its priority (1 high, 2 medium, 3 low), a one-line reason, and whether it is already installed or present. Then ask once, with AskUserQuestion multi-select, which to apply.

Composio: if ~/.dts2/composio-tools.json exists (the Composio tool map), improve-scan.py already leaves out MCP recommendations that a connected Composio toolkit covers and lists them under info as "covered by Composio ()". Prefer Composio toolkits over installing duplicate MCP servers or plugins, and point the user to /dt:composio to build or refresh the map.

4. Wire in

On the user's yes, handle only the picked items, per category:

  • skills: for each chosen skill:
    1. Install it if missing, with the publisher's installer for <source> --skill <name> (mattpocock skills use https://github.com/mattpocock/skills), into the project only; never -g. A failed install prints Skill <name> not installed (a package launcher unavailable or offline); continuing without it. and becomes a Decisions action item.
    2. Append it to .harness/keep.yaml under skills: as - name: <name> / source: <exact install command> / serves: harness, so a setup re-run keeps it. Append only; never remove entries.
    3. Wire it into the run workflow: companion skills follow the table under Companion skills in .harness/discipline.md; for a skill outside that table add a row naming when run should invoke it, and add it to the ## Skills table in .harness/end-state.md with its phase.
  • database: database tooling (db guard hooks, database MCP servers, deny rules for destructive commands, a migration check in CI, database skills). Never run migrations, resets, or destructive SQL. MCP servers are read-only by default, with credentials from env vars. Hooks go in through hooks-setup.py apply --ids db-guard,migration-check --dry-run first, then apply after the user confirms. Deny rules and CI follow the permissions and ci rules below.
  • hooks: run python3 -I "${CLAUDE_PLUGIN_ROOT}/scripts/hooks-setup.py" apply --ids <picked ids> --dry-run, show the result, then run the same command without --dry-run. It never overwrites existing hooks.
  • mcp: run claude mcp add --scope project ... only for the picked servers. Tokens go in environment variables the user sets; never type them into args or files.
  • permissions: show the exact lines first, back up .claude/settings.json, then append to permissions.allow or permissions.deny. Never remove entries. Remove a broad allow entry only when the user picks that item.
  • commands and subagents: write one short new file under .claude/commands or .claude/agents. Never overwrite an existing file.
  • ci, pr template, dependabot: copy the plugin template onto a new branch and hand off to /dt:ship. Never commit to main.
  • docs: run /init.

5. Report

One line per applied item, then a list of what was skipped.

Guardrails

  • Read-only until the user says yes.
  • Project scope only by default; user scope only if the user picks it.
  • Never remove or overwrite existing entries (keep list, hooks, permissions, commands, agents).
  • No secrets in files or commands.