Handoff
Skill stevennitesh/programming-agent-skills/skills/custom/handoff
Practical workflow skills for coding agents: scope, debugging, cleanup, subagents, GitHub tracking, and verification.
npx -y skills add stevennitesh/programming-agent-skills --skill handoffAssembled 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.
What its author says it does
Copied from the file, not written here
Compact the live thread into a resumable handoff for a fresh Codex session or agent thread.
SKILL.md
2.6 KB, 544 tokens by cl100k_base, as published. Nobody here has run it
Handoff
Trace -> Snapshot -> Compact -> Redact -> Save -> Verify -> Return
Trace. Resolve <work-root> as the Git root when present, otherwise the current directory. Target <work-root>/.tmp/handoff-<YYYYMMDD-HHMMSS>.md. In a Git repo, confirm the target is ignored before writing; otherwise recommend $repo-bootstrap and stop. Read the live thread and named artifacts.
Snapshot. Verify volatile repo and workflow state. Verify every pointer or mark it explicitly unverified. Label facts, inferences, and unknowns; leave new evidence and task work to the receiving session.
Compact. Preserve only state expensive to recover from the Source Trace, using the active workflow's vocabulary. A supplied focus sets Purpose and Next Step without hiding any blocker, risk, unresolved decision, or state needed to resume safely.
# Handoff
## Purpose
Continuation, decision, or proof target.
## Current State
Complete, in-progress, intentionally unchanged, and blocked state; active workflow and exact phase or gate. For repo work: cwd/worktree, branch or detached HEAD, relevant commit, staged/unstaged scope, material untracked files, and unrelated-dirty-work ownership.
## Key Decisions
Confirmed and rejected decisions, constraints, scope boundaries, and commitments awaiting approval.
## Source Trace
Exact pointers to durable truth and intentionally preserved `.tmp/` or tracked `.scratch/` artifacts. Reference; do not copy.
## Validation
Commands/checks, outcomes, proof, skips, gaps, and residual risk.
## Open Questions
Question, known evidence, and decision unlocked.
## Next Step
One executable, workflow-native action with target and stopping point.
## Suggested Skills
Receiving-session skills with one-line reasons, or `None`.
Redact. Remove secrets, credentials, and personally identifiable information.
Save. Create .tmp/ when absent and write exactly the target artifact. The invocation authorizes only those changes. Keep the file outside the index and commits; leave tracked files, tracker state, Git state, the active workflow, and Codex tasks unchanged. Suggested skills remain unexecuted.
Verify. Reread until the file is source-traced, redacted, actionable, current, pointer-exact or explicitly unverified, ignored when Git applies, and the only authorized state change.
Return. Report the absolute path and:
Continue from
<absolute-path>. Read the handoff first, then execute its Next Step.
Append a redacted one-line form of any supplied focus to the pickup prompt.
What ships with it: 1 file
43 B alongside SKILL.md
agents/
- openai.yaml43 B