skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
anentrypoint/gm56 installs

gm-continue

Works out what is left in gm from its rows, plans the whole batch of actionable rows, hands the batch to the gm skill, and calls the gm skill again while actionable rows remain. Use to continue gm work from its rows. Read skill://gm-continue/SKILL.md with ReadMcpResourceTool (server gm), or run the gm-continue prompt, before every finish of a gm run; it fills the missing rows and writes the run's notes to AGENTS.md and the process skill.

How do I install this agent skill?

npx skills add https://github.com/anentrypoint/gm --skill gm-continue
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    This skill is a workflow manager designed to coordinate the 'gm' and 'wfgy-method' processes. It manages state persistence and recursion limits while ensuring all tasks are verified before completion.

  • Socketpass

    No alerts

  • Snykpass

    Risk: LOW · No issues

What does this agent skill do?

gm-continue

  1. Setup. Before any work starts, produce the batch plan. If gm.stat_file({ path: 'tools/gm-continue-setup.mjs' }) finds a checkout with that script, run the read-only tools/gm-continue-setup.mjs; it plans every open row, reports any row that is neither open nor resolved as a defect, and prints the plan. If no checkout with the script is available, plan from the row verbs: call gm.prd_list({ status: 'open' }) and gm.unknown_list({ status: 'open' }), print Batch plan: <count> open row(s), planned before any work starts. and one plan <id>: <title> line for every listed row, and use gm.prd_list({ status: 'all' }) and gm.unknown_list({ status: 'all' }) to report any row that is neither open nor resolved as a defect. Nothing is worked before the plan is printed. Independent items in the batch are worked in parallel. Read skill://gm/WORKFLOWS.md for the graph shapes; keep PRD creation and PRD draining running while the graph produces rows, so the queue stays drained while it fills.
  2. Build, then witness. The whole batch is built first, all steps, and then witnessed in one pass over all of them. The commits land after that pass. Each step's witness is recorded in the run log and cited in DESIGN.md. A step is not witnessed on its own during the build.
  3. Fill and write notes. Before the hand-off, make the rows match the work, then record what the run learned:
    • For each piece of the task's literal words that no open row covers, add a row with gm.prd_add. Check the open list by title first, so a row is not added twice. Run the real path the request names, read its output, and add a row for any mismatch.
    • A project fact that the troubleshooting found goes to AGENTS.md, as one line under "Facts found by troubleshooting", with the DESIGN.md section that records it. A process lesson or workflow generalization is codified as a skill, in the section it belongs to (skills/dev/SKILL.md for the project's own process), never left as a note. Follow the memory rules in skill://gm/SKILL.md: the full AGENTS.md audit comes with every edit to it, and the file stays under 30 KB.
    • No row is held. An open row is implemented and resolved with a witness, or, for a PRD the owner has dropped, deleted with gm.prd_delete({ id }). An open question is resolved, never deleted.
    • Change the files with Edit only, and commit them by explicit path with gm-tool ship.
  4. Hand off. The next step is the gm skill: read skill://gm/SKILL.md with ReadMcpResourceTool (server gm), with the setup's list and what is left as its task. Its task is the setup output and the open rows, and nothing else. Any PRD row still open at this point goes in the hand-off with it.
  5. Re-enter. After the gm skill returns, read skill://gm/SKILL.md again with ReadMcpResourceTool (server gm), with what is left as its task, while actionable items remain. This is an invariant of this skill. A pass that closes no row is not a stop: the next step is still the list of what is left, and the gm skill read with it. The loop ends only when every actionable row was attempted and ended resolved, or when a pass closes no row and nothing more can be closed: no open rows remain, each is resolved or a dropped PRD deleted, finish needs a witness, gaps optionally name what the run did not check, and an open row blocks finish. Any other actionable row is a defect, and the setup reports it.
  6. End. The end prints the handoff only after the witness pass has run. The handoff lists the rows the pass resolved, any defect the setup reported, HEAD and the served hashes, the open runs and processes, and the next actions in order. Print it even when the list is empty.

INVARIANT: always read skill://gm/SKILL.md with ReadMcpResourceTool (server gm) when the list is produced, in this same run, with the list and what is left as its task. The next step after the list is the gm skill, never a question to the user. Do not send finish while work remains.

The setup is read-only and stays read-only. It writes nothing, commits nothing and changes no served behaviour. The only verbs used are the row verbs, named as gm.<verb>: gm.prd_add, gm.prd_resolve, gm.prd_delete and gm.prd_list, and their gm.unknown_ twins. The row-verb plan uses gm.prd_list and gm.unknown_list only, so it writes nothing.

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/anentrypoint/gm/gm-continue">View gm-continue on skillZs</a>