Add /whats-next skill: decision-support for picking the next thing to take on

[SnowboardTechie]

Problem / Motivation

/plan-workday and /weekly-planning are routines — they help me pursue consistent goals on a cadence. They don't help when the question is "I have time and energy right now, but I'm not sure what to take on." That's a different shape: not "execute the plan," but "what should the next thing even be?"

That moment shows up often, and decision paralysis stalls me on it. The cost isn't picking the wrong thing — it's not picking anything and losing the window.

The skill is primarily aimed at life goals (where I'm not naturally crushing it the way I am at work), but life goals routinely include work goals, so the skill needs to handle both.

Proposed behavior

/whats-next runs an explicit, opt-in decision-support flow. Mode is decided by cwd:

• Project mode — invoked from inside a code repo. Recommendation stays scoped to that project. Sources are repo-side (.notes/, AGENTS.md, GitHub issues, commits).
• Second-brain mode — invoked from inside the personal vault (notes_root from ~/.claude/cairn/identity.md). Recommendation can span any life area or project. Sources are vault-side (weekly plan, daily plan, life-area notes, goal anchors).

Flow:

1. Gather context. Read sources appropriate to the mode (see Implementation hints).
2. Q&A on current state. Ask whatever the gathered context can't answer on its own — energy level, time available, time-of-day vs. cognitive peak, mood/blockers. Skip questions whose answers are already obvious from context (e.g., it's 8am and cognitive_peak is 07:30–10:00).
3. Single recommendation with rationale. One thing to take on next, plus why it's the right thing right now (matches energy, advances a goal, unblocks something else, etc.). Not a ranked list — the whole point is to break the paralysis, and three options reintroduces the problem.
4. Hand-off. Offer a concrete next action: open the issue / open the note / kick off /issue-work / drop into the relevant directory.
5. Capture the session. Write the Q&A + recommendation as an ephemeral note in the personal vault — cairn-style capture. The note is intentionally short-lived (burnable after ~a week) and serves two purposes: memory across same-day re-asks, and a record I can skim later for patterns.

The user always gets to say "no, give me another one" — the recommendation is a starting point, not a verdict. On a re-ask, the skill reads the just-written ephemeral note so it doesn't repeat itself.

If a user wants to force second-brain mode while in a code repo (e.g., "I'm at my desk but the question is about life"), they can pass extra prompt context at invocation. No dedicated flag needed.

Scope

In scope

• New skill at plugins/cairn-notes/skills/whats-next/SKILL.md.
• /whats-next slash command (explicit invocation only — no natural-language trigger phrases).
• Mode detection from cwd (under notes_root → second-brain; else inside a git repo → project; else ask).
• Project-mode sources: repo's .notes/ agent workspace, AGENTS.md, current daily plan if present, open GitHub/Forgejo issues filtered to the user (mine / assigned / mentioned), recent commits.
• Second-brain-mode sources: current weekly plan, current daily plan if present, life-area notes, open goals, anything tagged or anchored as a goal/intention in the vault.
• Q&A for missing context (energy, time available, current cognitive-peak alignment).
• Single-recommendation output with rationale.
• Hand-off action where applicable (/issue-work, open file, etc.).
• Ephemeral session capture under the personal vault, in a folder understood to be burnable on ~weekly cycle.
• On re-ask within the same day, read the ephemeral capture so the second pick differs from the first.

Out of scope

• Modifying or replacing /plan-workday or /weekly-planning. This is additive — the routines stay routines.
• Extending forge to handle life-goal scope. forge stays focused on daily-goal sequencing; /whats-next owns its own synthesis logic to avoid bloating forge into something less useful for its current job.
• Natural-language trigger phrases ("I'm stuck", "what should I work on") — explicit invocation only.
• Dedicated mode-override flag. Extra prompt context at invocation is enough.
• Ranked top-N output. Single recommendation by design.
• Calendar integration in the first cut.
• Cross-project ranking inside project mode — project mode stays project-scoped.
• Persisting the recommendation to a daily or weekly plan file. The day's plan is owned by /plan-workday; this skill suggests, it doesn't author plans.
• Long-lived recommendation history. Ephemeral capture only; no permanent log.

Implementation hints

• Mode detection. Resolve notes_root from ~/.claude/cairn/identity.md. If cwd is under notes_root → second-brain mode. Else if cwd is inside a git repo → project mode. Else ask.
• Context gathering is the load-bearing step. Recommendation quality is bounded by what the skill reads. Worth being thorough here even if it makes invocation slow.
• Project-mode sources live in the repo, not the vault. This repo's whole convention is that .notes/ is the agent workspace — that's where project context lives. Don't go looking in ~/notes/second-brain for project signal during project mode.
• Source list overlaps with workday-planning. Look at how workday-planning reads daily-plan / GitHub / repo context and reuse the shape. Synthesis logic stays here, not in forge.
• Q&A should be sparse. Ask only the things gathered context can't answer. Time-of-day, working-hours, and cognitive_peak come from identity.md and the system clock — don't re-ask. Energy and mood are the things only the user knows.
• Synthesis logic is in this skill, not delegated to forge. Forge handles priority/dependency/energy for daily work goals. /whats-next reasons over a wider candidate set (life areas + projects + goals) and a different question shape (what to take on, not how to sequence today). Sharing the reasoning would bloat forge and dilute its current job.
• Ephemeral capture location. Use a clearly-labeled folder under the personal vault (e.g., ~/notes/second-brain/whats-next/ or similar — pick a name on implementation). Filename includes date so it's easy to glance at and easy to burn. Capture the Q&A answers, the recommendation, and the rationale.
• Burn cadence. Notes are intended to live ~a week. The skill itself doesn't do the burning — that's a separate cleanup pass (manual or via pyre). The skill just writes; it doesn't have to manage retention.
• Identity dependency. Skill needs identity.md for cognitive-peak awareness and vault detection. If missing, prompt to run /cairn-setup rather than asking inline.
• Worktree-aware. Project-mode runs from worktrees often. Resolve trunk root for .notes/ access (see agent-workspace/SKILL.md).
• Override via prompt context. Document that the user can override mode by passing extra prompt context at invocation (e.g., /whats-next I'm in a code repo but want to think about life stuff). No dedicated flag.
• No hand-off coupling. The hand-off offer (open /issue-work, etc.) is a follow-up question, not a chained flow. User can always say no and just keep the recommendation.

Acceptance criteria

• plugins/cairn-notes/skills/whats-next/SKILL.md exists with a description that triggers only on explicit /whats-next invocation, not on natural-language phrasings.
• Skill detects project mode vs. second-brain mode from cwd against notes_root in ~/.claude/cairn/identity.md.
• In project mode, skill reads at minimum: repo's .notes/, AGENTS.md, current daily plan (if present), open GitHub/Forgejo issues filtered to user, recent commits.
• In second-brain mode, skill reads at minimum: current weekly plan, current daily plan (if present), life-area / goal notes from the personal vault.
• Skill skips Q&A questions whose answers are already present in identity.md or context (e.g., does not re-ask cognitive peak when system time is inside the configured window).
• Skill asks for energy level when not derivable from context.
• Skill outputs exactly one recommendation with rationale tying it to the gathered context.
• On re-ask, skill reads the same-day ephemeral capture and produces a different recommendation.
• Skill writes an ephemeral capture note to the personal vault under a clearly-labeled, burnable folder.
• Skill offers a concrete hand-off action where applicable (open file, kick off /issue-work, etc.) and exits cleanly when declined.
• Skill does not write to the daily plan, weekly plan, or any non-ephemeral vault file. Read-only with respect to plan and goal files.
• Skill does not modify or extend forge. Synthesis logic lives entirely in whats-next.
• Plugin manifest registers the new skill so /whats-next is invocable.