Agent-Skills.md

Agent Skills: Recommend the next action

Assess current project state and suggest the most valuable next action. The primary entry point for users who aren't sure what to do. Analyzes context, checks for issues, and recommends specific skills or actions.

UncategorizedID: braselog/researchassistant/next

Repository

braselog/researchassistant
braselog
2

Install this agent skill to your local

pnpm dlx add-skill https://github.com/braselog/researchAssistant/tree/HEAD/.github/skills/next

Skill Files

Browse the full folder contents for next.

Download Skill

Loading file tree…

.github/skills/next/SKILL.md

Skill Metadata

Name
next
Description
Assesses repository, task, decision, GitHub, meeting, calendar, and unprocessed context state to recommend the highest-value next actions. Use for /next, session starts, uncertainty about priorities, or after completing work.

Recommend the next action

Read relevant project state

Consult only what is needed:

  • ~/.researchAssistant/researcher_telos.md
  • .research/project_telos.md
  • .research/phase_checklist.md
  • recent entries in .research/logs/activity.md
  • tasks.md
  • active records in .research/decisions/
  • open GitHub issues and pull requests
  • git status and recent commits
  • the calendar when scheduling affects feasibility

Treat phases as guidance, not rigid gates.

Check unprocessed context first

Inspect .research/context/daily/ for daily context files.

  • If a file exists from a previous date, flag it before planning:

    Unprocessed project context remains from YYYY-MM-DD. Run /wrap_up to reconcile it before relying on the current task and decision state.

  • Do not process or archive the inbox during /next.

  • A file for the current date is expected during an active workday and should not block recommendations.

  • If stale context materially affects confidence, state which recommendations are provisional.

Assess project state

Look for material conditions such as:

  • unprocessed meeting actions or candidate decisions;
  • local tasks duplicated by, missing from, or inconsistent with GitHub issues;
  • completed repository work still marked open;
  • closed work whose acceptance criteria were not verified;
  • decisions not reflected in code, analysis, or documentation;
  • stale DVC outputs or downstream artifacts after code or parameter changes;
  • drift among code, methods, results, figures, captions, and the manuscript evidence map;
  • failed checks, unresolved blockers, and important deadlines;
  • new data, meeting audio, figures, or manuscript gaps relevant to an active aim;
  • overdue wrap-ups or weekly, monthly, or quarterly reviews.

Use sync-project-state or project-health-check when inconsistency is material. Raise only issues that could change the recommended next action.

Priority order

  1. Correctness, blocked work, and unresolved scientific decisions
  2. Hard deadlines and dependencies
  3. The active aim's highest-value work
  4. Stale tracking, reproducibility, or documentation risks
  5. Routine planning and maintenance

When scheduling or near-term feasibility matters, use the calendar skill to read the relevant range with --brief. Do not query the calendar for purely technical prioritization questions.

Output

Provide a concise context summary and at most three options:

**Context:** [current aim/phase, recent progress, and any material warning]

A) **[Highest-value action]**
   [Why it matters and the concrete file, issue, decision, skill, or command]

B) **[Second action]**
   [Why]

C) **[Optional maintenance action]**
   [Why]

Mention uncertainty rather than pretending project state is fully synchronized. Do not print the full skill registry.