A collection of skills and tools for working with AI coding agents (pi / Claude). Each subdirectory is self-contained — own docs, own scripts.
| Directory | What it is |
|---|---|
ai-judge/ |
LLM-as-a-judge for voice-AI call transcripts: a deterministic pre-pass that catches "false success" defects by cross-referencing agent speech against the system action log, plus an LLM judge that scores the call against a rubric with verbatim-quoted evidence. |
resume-tailoring/ |
Workflow for tailoring a master .docx resume to a specific job posting or recruiter message. Edits the Word XML in place so fonts, styles, numbering, and hyperlinks survive; compression is subtractive (cut oldest roles first, never the most recent). |
writing-skills/ |
Guidance for authoring, testing, and deploying SKILL.md files — TDD applied to process documentation (write a pressure scenario, watch an agent fail without the skill, write the skill, verify compliance). |
code-simplicity-reviewer/ |
A skill that reviews code through a minimalism/YAGNI lens: flags unnecessary complexity, redundant code, and premature abstractions. |
improve/ |
Self-improvement workflow that mines a session transcript for friction (clarifications, repetition, gaps in the skill file), turns it into evidence-backed suggestions, and implements approved ones in an isolated git worktree across four approval gates — including improving itself. |
p2p-qa-lab/ |
Agentic QA lab for a Purchase-to-Pay API: an explorer that autonomously drives a vendor→PO→receipt→invoice→match→approve workflow, an adversarial layer that probes financial and security invariants, and a judge that emits auditor-style reports (deterministic re-computation anchored, LLM-narrated). |
Each component carries its own deep docs — start there:
- ai-judge →
ai-judge/README.md— input transcript format, prerequisites, quick start, judge flags, pre-pass checks, output schema - resume-tailoring →
resume-tailoring/README.md— directory layout, requirements, quick start, resume-format assumptions - writing-skills →
writing-skills/SKILL.md - code-simplicity-reviewer →
code-simplicity-reviewer/SKILL.md - improve →
improve/README.md— the four hard stops, two-model split, worktree isolation, checkpoint commands - p2p-qa-lab →
p2p-qa-lab/README.md— quick start, commands, architecture, report schema, tests
This is a public repo. Personal and interview-derived assets are never
versioned — they stay local via .gitignore:
*.docx/*.pdf(resume masters, LinkedIn exports)findings//judgments/(local AI-judge transcript analyses)runs//reports/(p2p-qa-lab demo output — report artifacts stay local)
A clone contains only code, docs, and skills.
GNU General Public License v2 — see LICENSE.