Pathfinder

Skill & Agent Router

Type /pathfinder when you need to know which skill or subagent to reach for. Nothing here fires automatically; this is a navigation aid only.

---

Session lifecycle

Start of session   →  /resume
                         ↓ (do work)
End of session     →  /handoff           writes .ai/HANDOFF.md, then calls /evolve-claude-md
Standalone update  →  /evolve-claude-md  promote session knowledge to CLAUDE.md only

---

Skills — user-invoked (type the slash command explicitly)

| Skill | Command | Reach for it when… | |---|---|---| | resume | /resume | Starting a new session; reconstructs context from .ai/HANDOFF.md | | handoff | /handoff | Ending a session or switching agents; calls evolve-claude-md internally | | evolve-claude-md | /evolve-claude-md | Standalone CLAUDE.md update without a full handoff | | grill-me | /grill-me | Stress-testing a plan or design before implementing; thin launcher for the grilling core | | slack-message | /slack-message | Drafting an internal Slack update grounded in git context | | tab-setup | /tab-setup [all] | Naming / coloring this Claude Code tab; all recolors every active session | | write-new-skill | /write-new-skill | Scaffolding a new skill from scratch | | ai-review | /ai-review | Full comprehensive repo/project review; orchestrates code-review/security-review/unstale/overbaked/reviewer-2 in parallel and adds gap-hunting + grounded ideation + prioritized synthesis. Run on fable at high+ effort | | pathfinder | /pathfinder | This router |

Skills — model-invokable (reachable by the model or another skill mid-task)

| Skill | Reach for it when… | |---|---| | grilling | Stress-testing a plan one decision at a time — and proactively before a step that changes running state, grabs a shared resource (GPUs/ports/memory), or could invalidate an earlier decision. /grill-me is the explicit launcher | | figure-review | User shares or references a figure/plot and wants it checked for publication readiness | | lit-review | Searching or synthesizing scientific literature; citation checks from a review flow | | overbaked | Auditing any artifact for over-engineering, verbosity, or scope creep | | reviewer-2 | Quick generalist stress-test of a claim, result, or section (any domain) | | unstale | Cleaning up dead imports, stale comments, resolved TODOs after a refactor | | evolve-claude-md | Promoting session knowledge to CLAUDE.md (called by handoff; also invokable standalone) |

---

Subagents — domain reviewer panel

Subagents are spawned with the Agent tool (subagent_type:), not slash commands. All four are review-only (report findings; never rewrite).

| Subagent | Domain | Reach for it when… | |---|---|---| | attribution-reviewer | Climate attribution | An attribution claim, result, or draft section needs expert scrutiny (PR/OR/ChIP, storyline, D&A) | | meteo-reviewer | Meteorology (AMS CCM) | A weather analysis or mechanistic argument needs dynamical / thermodynamic review | | stats-reviewer | Statistics / ML | A statistical analysis, methods section, or quantitative result needs estimator/causal/inference review | | scicomm-reviewer | Science communication | A public-facing product (article, press release, talk abstract) needs communication-effectiveness review |

---

Review decision tree

Need a review?
│
├── Quick generalist stress-test — one skill, any domain, fast
│   → /reviewer-2
│     Baseline, counterfactual, alternatives, uncertainty consistency.
│     No domain expertise assumed.
│
├── Deep single-domain review — expert read on one dimension
│   ├── Attribution claim / result     → attribution-reviewer subagent
│   ├── Weather / synoptic mechanism   → meteo-reviewer subagent
│   ├── Statistical analysis / ML      → stats-reviewer subagent
│   └── Public-facing science product  → scicomm-reviewer subagent
│
└── Full multi-lens review — manuscript, full section, ≥2 dimensions
    → Spawn the relevant subset of the 4 subagents in a SINGLE message
      (parallel, independent — no cross-anchoring).
      Synthesize findings after all return.
      Use /reviewer-2 first if you want a fast generalist pass before the panel.

When NOT to use the panel

• Single-domain question → one reviewer, not all four.
• /reviewer-2 covers the generalist case; don't over-convene the panel for a claim check.
• scicomm-reviewer is for public-facing products, not internal methods or data.

Citing literature during a review

Any reviewer can flag missing citations. To retrieve them, route to /lit-review (Zotero, arxiv, bioRxiv, Google Scholar, Consensus).

---

Composition rule (for skills that call other skills)

A user-invoked skill may call model-invokable skills; never another user-invoked one. (Examples: handoff → evolve-claude-md; grill-me → grilling.)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: dgilford
• Source: dgilford/ai-science-toolkit
• License: MIT

Install and usage instructions live in the source repository linked above.