Personal Agent Skills

Canonical portable skills for Neha's personal agents.

For the focused Brain onboarding bundle (Codex or any stdio MCP client), use the Brain install/preview flow. Version 0.2.9 is journal-first: it exposes canonical Brain reads plus provider journal.append, blocks ordinary-provider direct Brain mutations, includes exact frozen contract fixtures, and keeps published journal projections read only. It also includes doctor and reversible upgrade/uninstall. The all-skills installer below remains available for the broader personal-agent collection.

Install locally

The repository is private. The skills CLI supports private repositories using existing Git credentials, GitHub CLI auth, or SSH.

Codex + Claude Code + Cursor

After cloning:

git clone [email protected]:neha-nupoor/personal-agent-skill.git
cd personal-agent-skill
sh scripts/install.sh

Or directly:

npx skills add neha-nupoor/personal-agent-skill -g --agent codex claude-code cursor --skill '*' -y

The canonical skill tree is shared; do not maintain manual copies for each agent unless a real incompatibility requires an adapter.

Architecture

• skills/ — canonical skill definitions
• agents/ — portable identity and skill assignments; runtimes consume these but do not own them
• ROUTING.md — precedence and collision rules
• evals/ — behavioral regression cases
• scripts/ — local install and validation helpers
• personal-memory-service — separate shared state/provenance layer
• MCPs/plugins — capabilities such as Linear and personal memory

Skills

• ignition
• i-have-adhd
• active-plate
• rescue
• shutdown
• food-copilot
• research-to-decision
• scope-gate
• agent-team-architect
• conversation-to-linear
• context-pack
• memory-curator
• content-engine
• skill-gardener
• skill-scout
• bug-handoff
• momentum-guard
• sticker
• brain

Agent profiles

• personal — general personal assistant
• researcher — evidence-first research and synthesis
• chief-of-staff — ADHD-friendly priorities, open loops, and next actions

Hermes is the initial runtime. Cloud memory is canonical; Hermes-local memory is scratch state and may be deleted without deleting durable identity or memory.

Quality bar

Substantial skills should define trigger and non-trigger behavior, decision procedure/state transitions, collision/handoff rules, tool/degraded modes, user-visible interaction contract, success criteria, and regression cases.

Short skills are acceptable only when the behavior itself is genuinely small.

Validate

python3 scripts/validate-skills.py

Third-party

skills/i-have-adhd is the canonical upstream behavior from ayghri/i-have-adhd, MIT licensed. See third_party/i-have-adhd-LICENSE.