agentsclimarketplace

Harness update

Skill atlax-tech/harness-armor/skills/harness-update

Self-improving, tool-first Harness Engineering Agent Skills for Claude Code, Codex, Cursor and TRAE. Build, audit, update and verify AGENTS.md-based repositories.

Install
npx -y skills add atlax-tech/harness-armor --skill harness-update

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 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

Detect evidence-backed drift between a managed repository's product sources, architecture, implementation, commands, tests, and Harness; produce a file-level sync plan and apply it only after explicit approval. Use when a Harness Armor managed repository changed or synchronization is requested. Do not use for dependency updates, unmanaged Harness audits, or any silent overwrite.

The file declares its own license as CC-BY-NC-4.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.5 KB, as published. Nobody here has run it

Update a managed Harness

Work in two phases: immutable read-only proposal, then an explicitly authorized application. Invocation alone never authorizes writes.

Inputs

  • Repository root with a valid .harness/manifest.json.
  • Optional scope or suspected change.
  • A separate approval after the file-level proposal for any write phase.

Load the contract

  1. Treat this file's directory as SKILL_ROOT.
  2. Read references/authorization-and-drift.md.
  3. Resolve shared resources from SKILL_ROOT/../../shared or SKILL_ROOT/../.harness-armor.
  4. Read spec/harness-engineering-v1.md and the state schemas. Stop if missing.

Phase A — read-only proposal

  1. Read applicable AGENTS.md files and .harness/manifest.json.
  2. Run python scripts/validate_manifest.py <root>. If invalid, remain read-only and route to harness-check; do not repair state implicitly.
  3. Run python scripts/detect_drift.py <root>.
  4. Read every changed source and the affected code/config/test context. Hash differences prove change, not meaning.
  5. Detect and classify:
    • product or design source changes;
    • current architecture/module changes;
    • build/test/CI command changes;
    • Harness content drift;
    • invalid or obsolete rules;
    • user edits to managed content;
    • missing, added, or deleted evidence.
  6. Build assets/update-plan.md with evidence, affected files, proposed diff by file, ownership, risk, validations, and rollback.
  7. Separate NO_CHANGE, SAFE_MANAGED_UPDATE, USER_EDIT_CONFLICT, SOURCE_CONFLICT, and MANUAL_DECISION items.
  8. Stop and request explicit approval naming the plan version or exact files. Do not update timestamps, manifests, logs, or generated content before it.

Authorization gate

Valid approval must clearly refer to the displayed plan. Scope it to the named files and changes. General conversation, the original invocation, or approval of one file does not authorize all proposed files. If evidence changes after the plan, invalidate approval and regenerate the proposal.

Phase B — authorized application

  1. Re-check applicable files and hashes immediately before editing.
  2. Apply only approved changes. Preserve user-owned content and edit only authorized managed sections when ownership is partial.
  3. Never resolve a USER_EDIT_CONFLICT or SOURCE_CONFLICT unless the user's approval states the resolution.
  4. Update Harness documents, then source fingerprints and manifest baselines.
  5. Add a Chinese development log containing actual validations and manual acceptance steps.
  6. Run python scripts/validate_harness_structure.py <root> plus the approved project checks.
  7. Re-run drift detection. Remaining drift must be explained, not hidden.
  8. On validation failure, revert only edits made by this phase when their pre-edit bytes are known and unchanged by others. Otherwise stop and report the precise partial state.

Allowed and forbidden changes

Allowed only after approval: named Harness-owned files/sections, state fingerprints, and a development log. Business code can be read but not modified.

Forbidden: unapproved writes, silent overwrite, requirement reduction, source document rewriting, automatic refactoring, unknown script execution, or claiming unrun tests passed.

Idempotency and failures

An unchanged managed repository returns NO_CHANGE and writes nothing. Report invalid manifests, scan truncation, unreadable sources, changed evidence after approval, concurrent edits, and incomplete validation. Use harness-check for diagnosis when safe application is not possible.

Result format

Phase A reports the immutable plan and approval request. Phase B reports the approved scope, actual diffs, preserved conflicts, manifest/log updates, command evidence, residual drift, rollback status, and manual acceptance steps.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.