Claude Code skills I use day to day:
expert-opinion (direct, research-grounded decision consulting)
publicize (ship a private repo as public without breaking the version you still use)
clickup-sync (read-only sync of every ClickUp task assigned to you, zero config beyond the API token)
create-monday-task (structured Monday.com items with subitems and plain-prose context notes)
Distributed as a Claude Code plugin marketplace.
In Claude Code:
/plugin marketplace add mumtazm1/claude-code-skills
/plugin install claude-code-skills@claude-code-skills
Update later with:
/plugin marketplace update claude-code-skills
If you'd rather drop the skills straight into your user config:
git clone https://github.com/mumtazm1/claude-code-skills.git
cp -r claude-code-skills/skills/expert-opinion ~/.claude/skills/
cp -r claude-code-skills/skills/publicize ~/.claude/skills/
cp -r claude-code-skills/skills/clickup-sync ~/.claude/skills/
cp -r claude-code-skills/skills/create-monday-task ~/.claude/skills/Restart Claude Code. All four skills will register automatically.
Claude's default voice at decision points is too agreeable and relies on training-data priors. This skill makes Claude answer like a specific professional (a principal engineer, a fee-only CFP, an evidence-based PCP, and so on), pull fresh information from the web, and commit to a pick instead of listing tradeoffs.
Trigger it by saying "expert opinion", "second opinion", "what would a [role] say", or "help me decide". The skill enforces a four-step protocol: name the professional, pull current sources (with cross-referencing and date-stamping), synthesize internally, and return a short verdict with inline citations and a data-freshness line.
I asked Claude Code how I'd recommend syncing Postgres to Snowflake for a startup with ~100GB of data, as if I were the consultant answering.
First run, no skill, Claude gave me specific numbers. "4-6 hours to stand up the first source, ~1 hour per additional source." "Don't backfill through dlt or Airbyte, dump to Parquet first." It sounded authoritative.
I pushed back and told Claude to actually verify. It admitted the hours were fabricated and the dogma about backfills had no source. It was predicting plausible tokens instead of answering the question.
Second run, with /expert-opinion:
Pick: Airbyte OSS, self-hosted. Long-term optionality decides this. Airbyte has a published Python connector SDK and 300+ pre-built sources. PeerDB has three sources and no custom-connector framework. [PeerDB supported-connectors matrix, as of 2026-04]
Same question. Actual web searches, real connector matrices, inline citations, a data-freshness line. The answer I'd have written myself.
By forcing Claude to look things up instead of guessing, the results are much more grounded in reality.
See skills/expert-opinion/SKILL.md.
Turns a private repo into a public one without the usual failure modes: committed secrets surviving in history, commit dates getting nuked by --reset-author, drift between the private copy and the public snapshot, or README voice that reads like an AI wrote it.
The skill runs a nine-phase workflow (recover originals, sandbox, scan, triage, refactor-first, history rewrite, verify, README review, push private, review rendering, flip public) with hard gates at every destructive step. Uses gitleaks and git-filter-repo where available; falls back to ripgrep patterns when not.
See skills/publicize/SKILL.md and its supporting files (workflow.md, scanner-patterns.md, abstraction-recipes.md, history-rewrite.md, per-repo-archetypes.md).
Show every ClickUp task assigned to you across every workspace your API token has access to. Auto-discovers your user ID and workspaces from the token, so the only thing to configure is the token itself. Read-only, and API calls happen inside subagents so raw JSON never lands in your main context.
Trigger it by saying "clickup sync", "what am I tagged in on clickup", "check my clickup", or similar. Output is one compact table per workspace (task ID, status, due, priority, list, name), followed by pending/overdue items and recent comment activity grouped by task.
Needs one env var: CLICKUP_API_TOKEN. Ships a small clickup.py helper (Python standard library, no pip install). The skill uses ClickUp's "Get Filtered Team Tasks" endpoint with the assignee filter, so you don't have to list list IDs or workspace IDs manually.
See skills/clickup-sync/SKILL.md.
Creates a Monday.com item with a short deliverable-focused title, an HTML context note explaining what it is and why it matters, and subitems that break down the actual work. Each subitem gets its own one-sentence description. Shows a full preview before creating anything; nothing is written to Monday until you confirm.
Writes notes in plain, specific prose. No corporate filler, no "this initiative aims to streamline", no AI voice. The title and subitem style rules reflect my own preference for scannable boards; adjust them before using at team scale.
Needs a Monday.com MCP server configured for your workspace and one config value: the board ID. Group, priority, and status can be referenced by label name; the skill queries the board structure via the MCP and resolves labels to column/option IDs at runtime.
See skills/create-monday-task/SKILL.md.
MIT. See LICENSE.