Replan
Skill honerlaw/agent-marketplace/plugins/minerva/skills/replan
Records a course-correction for the current minerva work unit — drafts Original plan / What changed / New plan, stress-tests it via `minerva:grill-plan`, then appends a dated divergence entry to `.minerva/work/NNN-slug/replan.md`. Use when work has diverged from the proposal in a load-bearing way — a core assumption was wrong, the approach is changing, or scope is shifting — or to amend an approved proposal before `minerva:work` starts (pre-work tweaks), or when the user invokes `minerva:replan`.From its SKILL.md
npx -y skills add honerlaw/agent-marketplace --skill replanAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.7 KB, ~1.3k tokens by cl100k_base, as published. Nobody here has run it
Append a dated replan entry to the current work unit when reality has diverged from the proposal, OR amend an approved proposal pre-work.
Usage
minerva:replan— operates on the work unit inferred from current-session context, or the most-recently-modified if context is ambiguousminerva:replan 005-add-payments— operate on the named unit explicitly (slug or path)
When to use
- Mid-work divergence (most common): a core assumption broke, the approach is shifting, scope changed.
minerva:workauto-triggers this protocol. - Pre-work amendment: the proposal was approved but the user wants to tweak it before implementation starts. Same protocol — the replan entry's
Original planis just the freshly written proposal, andWhat changedis "before implementation: user wants to adjust X".
Target resolution
Same pattern used by minerva:work, minerva:promote, minerva:review, minerva:ship, minerva:cleanup. Keep all six blocks in sync if you edit one.
- Explicit argument — if the user passed a slug or path, resolve it. Look in both
.minerva/work/<NNN-slug>/and.minerva/worktrees/<NNN-slug>/.minerva/work/<NNN-slug>/. - Current-session context — if a unit slug, path, or branch name has been mentioned in this session, use it.
- Most-recently-modified across both locations — scan
.minerva/work/NNN-*/AND.minerva/worktrees/NNN-*/.minerva/work/NNN-*/, pick by directory mtime. - Ambiguity — list candidates, ask the user.
- None found — "no work units found — run
minerva:proposefirst" and stop.
Worktree addressing
After resolving the target and before reading or writing any files:
- Do not call
EnterWorktree— minerva worktrees live under.minerva/worktrees/, which that tool does not reliably enter; the session's working directory stays the parent repo. - If the resolved target's docs live at
.minerva/worktrees/<NNN-slug>/.minerva/work/<NNN-slug>/, address the worktree explicitly: prefix every file path this skill reads or writes with.minerva/worktrees/<NNN-slug>/, and run every git command asgit -C .minerva/worktrees/<NNN-slug> …. Relative paths resolve to the parent repo and silently misroute edits onto the wrong branch (see.minerva/knowledge/008-constraint-enter-worktree-absolute-paths.md). - If the docs live only on the default branch (a shipped unit being inspected), operate on the parent repo directly.
This keeps the skill correct regardless of where it's invoked from — every file path it touches names the right working tree explicitly.
Protocol
Same brainstorming pattern as minerva:propose, but framed around divergence:
- Read the existing context first. Read
proposal.md, any priorreplan.mdentries, and the currentscratchpad.md. The brainstorm must be grounded in what actually happened. - Frame the replan around three pieces:
- Original plan — what the proposal (or latest prior replan) said the approach was
- What changed — what was discovered, what broke, what assumption was wrong (for pre-work amendments: "user requested adjustment before implementation")
- New plan — the revised approach
- Ask clarifying questions one at a time to fill in any of the three pieces that aren't already obvious from the conversation or files.
- Propose 2–3 alternative new plans if the path forward isn't already settled. Iterate.
- Draft the replan entry internally, then stress-test it before showing it for approval. Assemble the Original plan / What changed / New plan triple in conversation (do not write to disk yet) and invoke the
minerva:grill-planskill via theSkilltool against it (no argument needed — it reads the draft from conversation). Let grill-plan walk the decision tree, edit affected pieces in place as answers surface, and return only once shared understanding is reached. The entry that exits grilling is what step 6 presents. - Present the resulting entry for approval before writing.
- Hard gate: do not append to the file until the user has approved the entry.
On approval — file write
-
If
replan.mddoesn't exist yet in the target work unit, create it with this header:# Replan log: <slug> -
Append a new entry using this exact template (today's date in
YYYY-MM-DD):## YYYY-MM-DD — <short, declarative title> **Original plan**: <one or two sentences> **What changed**: <what was discovered, what broke, what assumption was wrong> **New plan**: <one or two sentences> -
Update Success criteria if the new plan changes them. If the replan changes what "done" looks like, also edit
proposal.md's## Success criteriasection to reflect the new bar. Otherwise leave it alone — the replan entry will still supersede the proposal on conflict. -
Report the path and the title of the appended entry. Suggest resuming
minerva:worknext.
Out of scope
This skill stops at appending to replan.md (and optionally editing proposal.md's Success criteria). It does not invoke implementation — return control to minerva:work (or its in-progress session) after writing.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.