holycow-guide
Find your next step.
Tell it where you are. Get a useful next step without memorizing the toolkit.
When to use it
You have a task, but you are unsure which skill will help or whether you need a workflow at all.
What you get
A recommendation, a reason, and the result to expect. Guide reads the relevant skill definitions before recommending a route.
Try it in your agent
Invoke this in the agent chat after installing HolyCow.
/holycow-guide We have a new feature idea, but several product decisions are still open. Where should we start?$holycow-guide We have a new feature idea, but several product decisions are still open. Where should we start?HolyCow Guide
Recommend the smallest useful next step from the user’s current situation. Help them use the toolkit without memorizing it or following a mandatory sequence.
Establish the situation
Distinguish choosing a skill for project work from deciding what to add to the toolkit itself. Reuse the stated outcome, accepted decisions, completed work, and actual blockers. Ask a concise clarifying question only when missing information changes the recommendation.
Use the table below to shortlist candidates, then read each candidate’s SKILL.md before recommending it, including any offered alternative. Reuse a definition already loaded in this conversation unless it has changed. Check availability in the intended harness; a directory listing or catalogue description does not verify the skill’s behavior or boundaries. If a definition cannot be read, disclose that limit instead of describing its capabilities as verified.
| Immediate need | Candidate |
|---|---|
| Initialize or refresh repository workflow configuration | holycow-init |
| Resolve consequential uncertainty across multiple sessions | hc-compass |
| Challenge an idea or settle requirements through discussion | hc-refine |
| Answer a question with a prototype, experiment, or diagram | hc-lab |
| Turn accepted understanding into a brief or delivery tickets | hc-plan |
| Implement agreed behavior or investigate and fix a failure | hc-build |
| Examine changes for defects and requirements or standards gaps | hc-review |
| Explain changes, evidence, and merge risks in a PR description | hc-pr |
| Improve future work using evidence from past sessions | hc-retro |
Recommend a bounded route
Lead with one recommendation and why it fits now. State the expected result and any material prerequisite. Include a short alternative only when a real trade-off warrants it; make any implementation alternative conditional on its requirements and prerequisites being settled.
Preserve the user’s established role and harness assignments. Recommend an existing suitable session when known, rather than asking the user to choose its role again. If that session is unavailable, explain the limitation and offer a fallback. Do not invent a role split when none is established.
Start from the current stage. Clear work can go directly to Build; reviewing an existing PR need not restart planning. Init is optional, and Compass is for uncertainty that benefits from a persistent map. A failed check calls for investigation, not automatically a retrospective.
For a longer flow, show only the steps justified by the task and stop at the requested boundary. Do not prescribe every skill, an automatic new session, context clearing, or delegation.
If the useful skill is missing, disclose that and offer a direct approach or installation guidance as appropriate. Do not silently install it or pretend it ran. If no skill fits, say so rather than forcing the task into the catalogue.
For toolkit additions, compare the missing capability with existing skills and references first. Recommend a separate skill only when it has a distinct, reusable responsibility; otherwise propose a focused extension or evaluation of the existing behavior.
Advice alone does not launch agents, modify files, or publish work. When the user also requests the recommended action, load and follow the available skill and carry out that action within the established scope.