resume check

    1

    Your agent checks the handoff note against the real repo before acting on it, so a stale "tests pass" or "already pushed" doesn't send it down the wrong path.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    resume check

    Example session with this skill installed

    I'm back. Pick up where we left off - progress.md has the notes.

    • Read your context and instructions
    • Compiled the resume check

    Agent runs python3 scripts/resume_evidence.py progress.md first. Relevant lines of its output:

    note last commit: 2026-11-14T19:02:11+09:00 | file mtime: 2026-11-14T19:02:40+09:00
    
    ## commits since note time (a lead, not proof)
      9be01d3 Revert "rate limiter"
    
    ## remote freshness
      last fetch: 2026-11-15T08:50:03
    
    ## commit hashes named in the note (remote branches as of last fetch)
      4f2a9c1  commit  on remote: origin/main
    
    ## paths named in the note (exists? modified after note time?)
      MISSING  src/limit.py
      exists   logs/load.txt  size=48211  mtime=2026-11-14T23:10:05  CHANGED after note
               last line: DONE 2026-11-14 23:10 requests=1200000 errors=0
    
    ## PID 41877 - is it the same process? compare its start time with the note
    $ ps -o pid,lstart,etime,args -p 41877   (rc=1)
      PID                  STARTED     ELAPSED COMMAND
    

    Then, before touching code:

    2 of 4 claims are stale, and the limiter was reverted this morning - the next step changes.

    Claim (quoted)VerdictEvidence
    "committed 4f2a9c1 and pushed"holds, as of last fetchgit branch -r --contains 4f2a9c1 -> origin/main; last fetch 2026-11-15 08:50 (.git/FETCH_HEAD) - newer remote changes would not show
    "rate limiter in src/limit.py"stalegit log --oneline --since="2026-11-14 19:02" -> 9be01d3 Revert "rate limiter"; ls src/limit.py -> no such file
    "all green"needs a runcode changed since the claim (the revert); pytest -q would settle it - not run
    "load test job, PID 41877"staletwo angles agree: ps -p 41877 -> no process, and logs/load.txt ends with the end marker DONE 2026-11-14 23:10 ... errors=0 - it finished, it did not crash

    Changes since the note: 9be01d3 Revert "rate limiter" (message: "breaks burst test on CI"); working
    tree clean. The note does not mention either.

    Next step: not the Redis backend. Proposed: find out why the limiter broke the burst test, fix it,
    then return to Redis. Before that, two optional commands you may want me to run:
    git fetch (refresh remote state) and pytest -q (settle the test claim). Shall I run them, and
    shall I update the three stale lines in progress.md?

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A handoff note, progress file or compaction summary describes the repo at the moment it was written. By the time you resume, a commit may have been reverted, a branch rebased, a background job finished or died. This skill makes the first step of every resume a short, read-only check of each claim in the note against the repo as it is now.

    What it does

    • Collects the evidence in one pass. scripts/resume_evidence.py (Python standard library, read-only) prints the branch, working tree, commits since the note, stashes and last fetch time, then the current state of every commit hash, path, PID and systemd unit the note names.
    • Gives every claim a verdict - holds, stale, unverified, or needs a run - with the command and its output as evidence.
    • Catches the changes the note does not mention, which are usually the most important finding.
    • Re-checks the next step. The note's plan is followed only if no stale claim undermines it; otherwise the agent proposes a corrected one.

    Six traps it guards against

    1. A file existing is not a job finishing - it looks for an end marker, not the file.
    2. A process search can match itself - it uses the PID and compares its start time.
    3. A log's modified time is not the event's time - it reads the time inside the line.
    4. A new commit on a shared branch says nothing about who made it.
    5. One empty look does not prove "did not happen" - it checks from two angles.
    6. When one piece of evidence falls, only that claim goes back to unverified.

    It also flags commits the note names that no branch reaches any more (rebased, amended or reset away) - git log still shows them, so they are easy to miss.

    What it will not do

    No git fetch, no tests or builds, no network calls (including gh), no edits. Claims that need one of those are marked "needs a run" and the agent asks before running anything. It edits your notes file only on your yes, and only the stale lines.

    Requirements

    Git and Python 3. Works with any agent that can run shell commands.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Workflow Automation.

    Read-only local checks only. No network operations (git fetch), test suite execution, or build runs are performed during the check. Requires a local git repository.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 2 days ago

    • Free with an Agensi account

    Software
    GitPython 3

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Shell runs read-only git, ps, ls and systemctl commands plus the bundled read-only Python script. Write Files is used only to fix stale lines in your notes file, and only after you say yes. No network access.

    Listed2 days ago

    What's inside

    Frequently Asked Questions

    The 'resume check' skill prevents agents from acting on outdated or hallucinated information by cross-referencing handoff notes with the live Git history and filesystem before starting work.

    Any agent that follows SKILL.md and can run shell commands in the repo: Claude Code, Codex CLI, Cursor, Gemini CLI and others. It needs only Git and Python 3; it does not use the GitHub CLI or any network call.

    To maintain safety and speed, the skill uses read-only commands like 'git cat-file' and 'stat'; it will not perform network-heavy actions like 'git fetch' or trigger long-running builds without your explicit permission.

    SKILL.md (when to run the check, the steps, six traps and the report format), scripts/resume_evidence.py (a read-only evidence collector) and one worked example of a handoff that went stale. It is free.

    Simply add the skill to your agent's toolkit; it automatically scans for common progress files in your root directory and begins the verification protocol whenever a handoff is detected.

    Only as far as your last git fetch, and it says so with the fetch time. It never fetches by itself; if the answer depends on the remote, it marks the claim needs a run and asks before fetching. Local commits since the note are always listed.

    New in Workflow Automation

    Free