agentleFS
akshay-diwadkarakshay-diwadkar/first-four-hours/skills/workbench/construct/SKILL.md

construct

Skill1 starsChanged 4 days agoMITView on GitHub

Implement an existing sealed `blueprint.md`; run its Verification; report completion, incomplete verification, or contradiction; skip planning-required work.

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. rtmandreyev/construct★ 134same name
    2. saleh-alhaddad/itqan-engineering★ 10same name
    3. maxgfr/construct★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository17 other files it gives its agents.

The file

Files
SKILL.md

Construct

Authority

Construct persists no execution state and creates no execution artifact. Blueprint owns every planning decision; Git worktree state owns the implementation delta. Discovery is evidence, not authority: it cannot expand a sealed Blueprint. The Blueprint stays byte-for-byte unchanged. Preserve unrelated dirty work; execution requires a Git worktree. Scripts prove mechanics only; the agent owns attribution, scope truth, and completion.

Command

python <construct>/scripts/cli.py --repo-root <repo> --input blueprint_file=<blueprint.md> --input phase=before --format json verify

Judge criteria first; follow next_command when present, else required_action.

Smooth execution loop

  1. Intake. Read Execution contract once, then run verify before for fresh work, status for resume. For dirty targets, non-Git paths, concurrent, destructive, or recovery work read Safety. Read the whole Blueprint before editing; when a caller quotes or paraphrases it, reread that step and work from its sealed text. Each turn re-reads your context, so issue independent reads and checks together, calling alone only when the input depends on an earlier result.
  2. Implement each ### N. in order. apply writes declared blocks, never by hand: with Red:, apply Frozen, write prose tests, prove its declared red, then apply Exact. Implement the rest, then prove green with the caller-held next unused attempt and follow its forward guidance. Repair only what Blueprint settles; otherwise stop as plan-contradiction.
  3. Verify. Run verify phase after; judge every Expect:. Execute Runs only through verify or reverify, on resume too. Repair locally only when Blueprint fixes the answer, then reverify from the earliest invalidated Run line. A shown declared Incomplete: outcome ends as verification-incomplete after step 4. Else stop with plan-contradiction evidence. Timeout or mutation failures are unsafe, never reverify.
  4. Reconcile and report. Compare the final diff against the sealed Blueprint once, per the Execution contract. Consequential behavior or safety risk, meaningful propagation, indirect verification, or a hard-to-observe acceptance condition earns an Earned Conformance Review; else none. Report instructions, each Proof run and Run with its Expect: and quoted output showing whether it met it, unrecoverable red gaps, notices, any caller quote the Blueprint lacks, whether a review was earned, run, or earned and not run, and whether protected paths and HEAD are unchanged.

Exceptions

Stopping ends the Construct run: plan-contradiction, protected-work-conflict, or safety-contradiction, reported with exact evidence. Failure, resume, and stopping owns failing Runs, resume, and what happens next.

Final and stop reports end with one optional next step fitting this outcome, after a stop its recovery owner's: a fenced, self-contained prompt naming any repository, issue, PR, or artifact it needs, or say none is needed. Only a "yes" starts it, via any skill it names, but the user runs a charter, diagnose, or ticket step as its /<skill> command.

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.