agentsclimarketplace
Sign in

Reversible agent task

Skill markoblogo/abvx-agent-skills/skills/reversible-agent-task

Run risky or multi-file agent work as a reviewable proposal before mutating the target workspace. Use when a task should produce retained output, inspectable diffs, and an explicit select/apply/discard decision instead of direct edits.From its SKILL.md

Install
npx -y skills add markoblogo/abvx-agent-skills --skill reversible-agent-task

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

One thing to look at

  • 14 stars14 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 file declares

Copied from the file, not written here

The file declares its own license as MIT. 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

3.3 KB, 619 tokens by cl100k_base, as published. Nobody here has run it

Reversible Agent Task

Use a proposal-first workflow when direct mutation would make review, rollback, or comparison harder.

Use When

  • an agent runner, SET bundle, or planning doc mentions retained output, proposal lifecycle, settlement, or reversible traces;
  • the task can be attempted in a worktree, draft branch, sandbox, retained-output folder, or external runner proposal;
  • multiple possible implementations should be compared before one is applied;
  • the target repo should remain unchanged until a human or supervisor reviews the output.

Lifecycle

  1. run: execute the task in a proposal context, such as a worktree, draft branch, sandbox, or retained-output directory.
  2. retained_output: keep generated files, diffs, logs, and proof artifacts separate from the target workspace.
  3. inspect: review diff scope, verification evidence, assumptions, and expected artifacts.
  4. select: mark one retained output as the candidate result.
  5. apply: merge or copy the selected proposal into the target workspace only after gates pass.
  6. discard: reject the proposal and leave the target workspace unchanged.

Procedure

  1. Define the target workspace and the proposal workspace.
  2. State the mutation rule: no direct writes to the target workspace until apply.
  3. Capture expected artifacts, verification commands, and retained-output path before the run.
  4. Run the agent or implementation loop in the proposal context.
  5. Inspect the retained output:
    • changed files;
    • generated artifacts;
    • tests or checks run;
    • unresolved assumptions;
    • scope drift.
  6. Choose exactly one settlement action: select, apply, or discard.

Output Shape

Include:

  • target workspace;
  • proposal workspace or retained-output location;
  • lifecycle state reached;
  • artifacts produced;
  • verification evidence;
  • settlement action: select, apply, or discard;
  • blockers or assumptions.

Guardrails

  • Do not present retained output as applied work.
  • Do not hide failed proposals; summarize why they were discarded.
  • Do not mutate the target workspace during run or inspect.
  • Do not use this skill for trivial one-file edits where normal git diff review is enough.
  • Keep the proposal mechanism host-agnostic: worktree, branch, sandbox, retained directory, or external runner are all acceptable.

Pair With

  • pipeline-readiness-gate for pre/post/ship gate selection.
  • assumption-excavation before running high-ambiguity proposals.
  • confidence-fragility-review before applying a proposal with public or release claims.
  • delivery-baseline-audit before calling applied work complete.
  • filesystem-context-discipline when retained outputs, scratchpads, or handoffs need durable file structure.

Provenance

Adapted from proposal lifecycle patterns in reversible agent runtimes such as shepherd-agents/shepherd, narrowed into a host-agnostic ABVX review discipline. This skill does not require Shepherd.

What ships with it: 2 files

1.7 KB alongside SKILL.md

agents/

Gives 0 of the 12 instructions most context ai engineering skills give in 619 tokens

Counted across 1,193 of the 1,976 authors here whose files we hold, read 2026-08-07

  • Dispatch a fresh implementer subagent per taskin 48 of 1193, across 19 files
  • Dispatch a final code reviewer after all tasksin 33 of 1193, across 8 files
  • Provide full task text to the subagentin 30 of 1193, across 9 files
  • Review spec compliance before code qualityin 27 of 1193, across 10 files
  • Make the hook script executablein 26 of 1193, across 8 files
  • Re-snapshot after navigation or DOM changesin 25 of 1193, across 19 files
  • Read files before editing themin 22 of 1193, across 11 files
  • Answer subagent questions before proceedingin 22 of 1193, across 7 files
  • Mark task complete in TodoWrite after approvalin 22 of 1193, across 6 files
  • Merge hook into existing settingsin 21 of 1193, across 3 files
  • Ask if installation is global or projectin 20 of 1193, across 2 files
  • Copy the hook script to target locationin 20 of 1193, across 2 files

Said here and by no other author read

  • Keep generated files separate from the target workspace
  • Inspect retained output before selecting a proposal
  • Choose exactly one settlement action
  • Merge proposals only after gates pass
  • State the target workspace mutation rule before running
  • Capture expected artifacts and verification commands before the run

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,144. 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.