Install
$ agentstack add skill-85danf-agent-skills-plan-handoff-prompt ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Plan Handoff Prompt
Generate a stand-alone kickoff prompt that a fresh Claude session can paste-and-run to continue plan-driven work without re-deriving context. The output is a prompt the user pastes into a new session, not something the current session executes.
When to Use
- The user says "prepare a prompt I can give you in a fresh session", "kickoff prompt", "handoff prompt", "next-session prompt", or "draft the prompt for next time".
- The current session has finished a phase (plan written; tasks 1–N committed) and a follow-up phase will run separately.
- A plan file (and often a design/notes/spec file) exists under
docs/superpowers/.
Required inputs (ask only for missing; never invent)
- Phase verb —
picking up,running,implementing,executing,validating, orcontinuing. - Working directory — absolute path. Verify with
git -C rev-parse --show-toplevel. - Current branch + main branch. Verify with
git -C branch --show-currentandgit -C symbolic-ref refs/remotes/origin/HEAD --short. Trust the user if the symref is unset. - Plan file path (relative to working dir). Required. Read first ~40 lines for task count and step style (
- [ ]checkbox vs free-form). - Companion design/notes/spec — optional. Include when present under
docs/superpowers/notes/or/specs/. Label by role: "canonical 'why' document", "design rationale", or "ground-truth spec". - Required state — commits or test counts next session confirms before starting. Get from
git log --oneline ..HEAD(last 10–20). Include commit-prefix convention (e.g.-) and the exactgit logcommand. - Required skill — usually
superpowers:subagent-driven-development(independent tasks, same session) orsuperpowers:executing-plans(parallel session). Match the plan's structure. - Friction notes — gotchas next session needs (stale .class files, IDE false positives, credential issues). Skip the section if none.
- Commit convention — JIRA prefix from branch name (
bugfix/-→-), per the user's CLAUDE.md.
Output structure (canonical section spine)
Read template.md (sibling to this file) for the full spine. It defines, in fixed order, eight ## sections — Working directory and branch, Required state, Read these files first / Plan file, Why this work exists, Required skill, Baseline check, Known friction, Commit convention — with placeholder hints for each. Emit sections in that order, drop empty ones, and use exactly one of the two file-block headings depending on inputs.
Drafting rules
- Backticks around paths, branches, commands, commit hashes.
- Absolute paths for working directory; relative paths (to working dir) for plan/notes/spec/test files.
- Quote next session's gates verbatim —
git log --oneline HEAD~10..HEAD,mvn -pl ingest test, etc. - No invented state. Didn't read the plan? Don't assert task count. Didn't run
git log? Don't list commits. - One "stop" condition per gate. Always include "If commits NOT present, STOP and ask…" on Required state.
- Drop empty sections.
- Preserve user's phrase bank when it fits: "do NOT batch tasks", "STOP and ask me", "spec-compliance reviewer + code-quality reviewer between tasks".
Workflow
- Read the plan file (always). Read the notes/spec file if cited (always). Run
git -C status -sb && git -C log --oneline -20against the working dir to ground Required state. Never let these commands fall back to the current shell CWD — that is the wrong-tree case the gate catches. - Ask only for inputs you cannot derive: phase verb, required skill (if ambiguous), friction notes, and any "Required state" details not in the commit log.
- Draft the prompt by filling the spine. Drop empty sections.
- Save to
~/.claude/tmp/next-session-prompt.md(creating the directory if needed) AND print it inside a fenced`markdownblock. - Tell the user: paste from chat, or run `pbcopy plan (code-changes phase) in using subagent-driven execution.
Working directory and branch
- Working directory:
/Users//work/ - Branch (already checked out):
bugfix/- - Main branch:
master - The branch is ~12 commits ahead of origin (prior-session work). New tasks build on top.
Plan file
docs/superpowers/plans/.md— N tasks with verbatim code blocks; steps use- [ ]checkboxes. Follow task-by-task. Do NOT touch the companion verification plan — that is a separate session.
Required skill
Use superpowers:subagent-driven-development. Dispatch a fresh subagent per task with the FULL task text + context inline. Spec-compliance reviewer + code-quality reviewer between tasks. Do NOT batch tasks.
Why this work exists
A scenario-validation pass surfaced three regressions in . These tasks plug each one without breaking the off-by-default behavior from the prior session.
Commit convention
Commit messages use the prefix - , matching what is already on the branch.
## Red flags — stop and re-draft
- Prompt has more than 8 `##` sections → drop empty placeholders.
- No backticked paths → will rot when files move.
- Asserts unverified test counts/commits/branch state → invented state. Re-run `git log`, read the plan, or ask.
- Required state lacks the "STOP and ask" gate → next session continues blindly on a wrong tree.
- Wrapped in extra commentary instead of a single fenced ```markdown block → the output IS the prompt.
- JIRA prefix does not match branch name → re-derive from `bugfix/-` → ` - `.
## Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- **Author:** [85danf](https://github.com/85danf)
- **Source:** [85danf/agent-skills](https://github.com/85danf/agent-skills)
- **License:** MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.