agentleFS
formal-youformal-you/agent-skills-suite/skills/workflow/converge/SKILL.md

converge

Skill0 starsChanged 28 days agoMITView on GitHub

Fold confirmed implementation deviations back into the technical spec after a task reaches QA_PASSED, archive the task bundle and close the change proposal. Use when a task bundle records a checked QA_PASSED verdict and its 02-dev-workspace.md deviation log is ready to be written back. Do not use while implementation or QA is still running, and do not use to invent spec content.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use --dry-run first to inspect the planned writes.

    line 62
  2. Distinctive:

    When the helper is unavailable, perform the same steps by hand and keep the markers exact:

    line 70
  3. Very specific:

    Read the deviation rows from 02.

    line 73
  4. Distinctive:

    Stop and report instead of converging when any of these is true:

    line 84

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Converge - Spec Write-Back After QA

Close the execution loop: a task that passed QA becomes part of the living specification. This skill represents the ConvergeTool role in the lifecycle. It defines the process; repository agent rules still own permissions, and this skill never grants extra access.

Pipeline position: dev -> qa -> converge -> archived bundle + updated spec.

Use when

  • A task bundle's 03-qa-verification.md records a checked QA_PASSED verdict.
  • 02-dev-workspace.md contains the frozen Implementation deviations table (rows or an explicit None row).
  • The task's confirmed behavior must be folded into .specs/02-tech-spec.md, the bundle archived and the originating change proposal closed.

Do not use when

  • Implementation or QA is still running, the handoff marker is unchecked, or REWORK_REQUIRED is active - wait for a stable QA_PASSED verdict.
  • The deviation log is blank or missing - send the task back to Dev instead of guessing.
  • You would have to invent spec content or rewrite existing spec prose; that is a planning change, not a convergence.

Inputs

  • tasks/<task-id>/01-contract.md - title, traceability anchors, optional change_id.
  • tasks/<task-id>/02-dev-workspace.md - the Implementation deviations table (frozen format).
  • tasks/<task-id>/03-qa-verification.md - the checked QA_PASSED verdict and evidence.
  • tasks/<task-id>/04-process.md - derived status card and event log.
  • .specs/02-tech-spec.md - the living technical specification.
  • .specs/changes/<change-id>/proposal.md - the originating change proposal, when one exists.

Outputs

  • One dated, attributed entry per task inside the spec's Converged changes section.
  • A convergence marker per task: <!-- converged: <task-id> sha256:<hash> -->.
  • 04-process.md with macro_status: "CONVERGED" and one TASK_CONVERGED event row (actor ConvergeTool).
  • The task bundle moved (never deleted) to tasks/archive/<task-id>/.
  • .specs/changes/<change-id>/convergence.md plus proposal status archived, when a proposal exists.

Check -> apply -> verify

  1. Check. Classify bundles before writing anything: python .agents/skills/converge/scripts/converge.py check --tasks-dir ./tasks --json A bundle is converge-ready only when QA passed, no rework is active, the deviation table has at least one row or an explicit None row, and the spec does not already carry its marker.
  2. Apply. Converge one bundle: python .agents/skills/converge/scripts/converge.py apply --task tasks/<task-id> Use --dry-run first to inspect the planned writes. The helper refuses with exit 1 when the bundle is not converge-ready, and refuses an existing archive destination unless --force.
  3. Verify. Re-run check, confirm the marker is present, 04-process.md says CONVERGED, the bundle is archived, and --spec-sha256 matches the expected CI value: python .agents/skills/converge/scripts/converge.py --spec .specs/02-tech-spec.md --spec-sha256

Manual fallback (no Python)

When the helper is unavailable, perform the same steps by hand and keep the markers exact:

  1. Confirm QA_PASSED is checked in 03 and no REWORK_REQUIRED is active.
  2. Read the deviation rows from 02; reject a blank log.
  3. Append (do not edit) one entry to the spec's Converged changes section: a ### <task-id> heading, the deviation table, and the <!-- converged: <task-id> sha256:<hash> --> marker. Compute the hash with any SHA-256 tool over the normalized task ID and rows.
  4. Set macro_status to CONVERGED in 04 and append the TASK_CONVERGED / ConvergeTool row.
  5. Move, never delete, the bundle to tasks/archive/<task-id>/; archive the proposal and write convergence.md when a change proposal exists.
  6. Re-run the verify step and record the evidence.

Refusal conditions

Stop and report instead of converging when any of these is true:

  • QA not passed - 03 has no checked QA_PASSED, or REWORK_REQUIRED is still active.
  • Unresolved rework - retry evidence shows the task is mid-rework; wait for a new verdict.
  • Blank deviation log - the Implementation deviations table is missing or has no DEV row and no explicit None row.
  • Spec drift - a modify/remove deviation points at a #sec-* anchor the spec does not contain; the spec and the deviation log disagree. Run check/apply with --strict in CI to turn this warning into a refusal.
  • Archive collision - tasks/archive/<task-id>/ already exists and --force was not given.

Structural deviations (new modules, changed boundaries, new ADR-level decisions) must escalate to spec-generator; they are never converged silently.

History and attribution

Convergence never rewrites history: it appends a dated, attributed section instead of editing existing spec prose. Existing entries stay byte-for-byte stable, and every entry names the source task, the QA verdict and the ConvergeTool actor. Never edit an old convergence entry to make a new task fit; add a new entry and record the conflict.

References

  • references/convergence-protocol.md - state machine, who writes what, archive layout, idempotency markers, conflict rules and structural escalation.
  • Role boundaries: references/role-boundaries.md — write scope, invariants and the honest enforcement model (bundled; no external files needed).
  • Task bundle format and role lifecycle: task-decomposition skill (when installed).
  • Handoff input: dev skill; verdict input: qa skill.
  • Repository rules: read AGENTS.md and any role-rules files when the adopting project defines them; they may add requirements but are not required for this skill.

Honest limits

This skill does not grant write access and cannot enforce role isolation. Where the repository has CI (for example a git-based role audit) or branch protection, that is the authoritative layer. The bundled helper is an optional deterministic assistant; the markers and the written evidence are the contract.

Image: untitled

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.