agentic-engineering-skills-personal

Private collection of reusable Claude Code skills, shared across projects.

Install

Clone once, then symlink each skill into the personal skills directory:

git clone [email protected]:Chasonnnn/agentic-engineering-skills-personal.git ~/agentic-engineering-skills-personal
ln -s ~/agentic-engineering-skills-personal/audit-skill ~/.claude/skills/audit-skill
ln -s ~/agentic-engineering-skills-personal/board-sync ~/.claude/skills/board-sync
ln -s ~/agentic-engineering-skills-personal/authmux ~/.claude/skills/authmux
ln -s ~/agentic-engineering-skills-personal/codex ~/.claude/skills/codex
ln -s ~/agentic-engineering-skills-personal/github-pr-validation-loop ~/.claude/skills/github-pr-validation-loop
ln -s ~/agentic-engineering-skills-personal/no-use-effect ~/.claude/skills/no-use-effect
ln -s ~/agentic-engineering-skills-personal/orchestrate ~/.claude/skills/orchestrate
ln -s ~/agentic-engineering-skills-personal/review-and-remediate-server-logs ~/.claude/skills/review-and-remediate-server-logs

Mods (Claude Code function-hook plugins) load from CLAUDE_CODE_PLUGIN_DIRS in the env block of ~/.claude/settings.json:

"CLAUDE_CODE_PLUGIN_DIRS": "~/agentic-engineering-skills-personal/mods/progress"

New machines: repeat both steps. Updating: git -C ~/agentic-engineering-skills-personal pull (symlinks pick up changes automatically).

Skills

| Skill | What it does | | --- | --- | | audit-skill | Audit or improve skill instructions and helpers with source-specific evidence; includes optional comparison guidance and the Provencher article reference. | | board-sync | Sync planning state (specs/ADRs/roadmap) to a GitHub Project board as epic issues with nested sub-issues — epics on the board, slices hidden under them — plus an optional stale-issue refresh. | | authmux | Resolve repository-bound authentication intent, verify provider identity evidence, route supported CLI commands through process-scoped contexts, and fail closed before wrong-account or wrong-project work. | | codex | Wraps the OpenAI Codex CLI for an independent second opinion: review (PASS/FAIL/INCOMPLETE on an explicit diff scope), challenge (adversarial break-it), consult (ask anything, with session continuity). Ported from gstack's /codex skill — see credit note below. | | dependency-modernization | Upgrade dependencies with official release review, local feature adoption, compatibility checks, and measured validation. | | github-pr-validation-loop | Audit open PRs against the current base branch, reimplement valid findings cleanly, close invalid or superseded noise with evidence, and monitor CI through completion. | | no-use-effect | Prefer explicit React data flow, replace ad-hoc Effects with derived state, event handlers, query libraries, or keyed remounting, and contain legitimate external synchronization in named hooks. | | orchestrate | Coordinates delegates, independent review, and integration; shares CLI execution with codex. (Folded in from orchestrate-skill with history; that repo is archived.) | | review-and-remediate-server-logs | Review live server logs, prioritize actionable failures, reproduce and fix their causes, validate through CI, and verify the result after an authorized deployment. |

Mods

| Mod | What it does | | --- | --- | | progress | Progress bars above the prompt: the mcp__progress__track tool, a hairline bar with stage notches and a moving shine (heavy-rule text bar in the terminal), elapsed time, subagent strips, a done line that fades out, a status-line fallback while hidden, and session names taken from the open bar or a one-time Haiku call. Commands: /bars, /bars-clear, /bars-demo. |

Conventions

• One directory per skill, SKILL.md inside, frontmatter per Claude Code skill format. Mods live under mods/, one plugin folder each.
• Skills here are project-agnostic: per-project facts (repo, board number, doc paths) are inputs, never hardcoded.
• Preserve each skill's invocation policy. Keep discovery concise, load conditional detail through references, and retain exact procedures where correctness requires them.

Credit

codex/ is a standalone port of the /codex skill from gstack (Garry Tan, MIT license) — the original wraps the OpenAI Codex CLI as one of ~30 skills in a much larger multi-agent toolkit. This port keeps the review/challenge/consult mechanics and drops everything wired into gstack's own infrastructure (telemetry, question-tuning auto-decide, checkpoint mode, gbrain sync, the D<N> AskUserQuestion format), so it runs standalone with no dependency on gstack being installed.

Helper validation

Use the approved project/host Python runtime. Helpers use the standard library.

claude plugin validate mods/progress && claude plugin test mods/progress
python -m unittest discover -s codex/tests -v
python -m unittest discover -s github-pr-validation-loop/tests -v

The PR snapshot JSON now contains coverage and pull_requests; exit 2 signals incomplete coverage. The CI watcher requires explicit workflow IDs and a full commit SHA. See its skill for event selection, authentication, and remaining readiness checks.