project-execution
Skill1 starsChanged 4 months agoMITView on GitHub
Drive a large, multi-stage project from goal to verified completion with persistent state, staged decomposition, task-by-task execution, integration checks, and honest evidence. Use for end-to-end projects, multi-session research, papers, reports, analyses, operations, codebases, or features that cannot be completed reliably in one pass. For software, applies a simple and surgical coding policy and can route to an already-installed GSD or Superpowers workflow. Do not use for a small standalone task, a factual answer, or planning when the user has not asked to execute.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- decision-plannerSkill
- diagrammingSkill
- project-one-sliderSkill
- project-understandingSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. About 0 are common, 15 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈15
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Run long work from files and evidence, not conversational memory.
- Distinctive: line 27
Read references/software-backends.md before delegating.
- Distinctive: line 35
Write a concrete goal, deliverables, exclusions, constraints, and acceptance checks.
- Distinctive: line 39
Create a small number of sequential stages such as research, design, build, integrate, verify, and polish.
- Very specific: line 49
Use the project's existing issue tracker or planning convention when one exists.
- Distinctive: line 51
Keep the master plan current.
- Distinctive: line 57
Read the goal, current stage summary, dependencies, and relevant conventions.
- Distinctive: line 60
Run its specified check.
- Distinctive: line 63
Update the stage summary and master plan before moving on.
- Distinctive: line 69
When assembling task outputs, audit them one by one:
- Distinctive: line 79
Use the strongest applicable checks: executable tests, recomputation, lint/build, schemas, invariants, boundary cases, source/citation …
- Distinctive: line 81
After any failure, fix it and rerun the full relevant suite.
- Distinctive: line 87
Every required task is verified or explicitly waived by the user.
- Distinctive: line 88
Every accepted task is incorporated.
- Distinctive: line 93
If pausing, leave a handoff containing current stage, last verified result, active blockers, exact next task, and commands or files needed …
- Distinctive: line 104
Do not expose the entire internal workspace unless requested.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Project Execution
Run long work from files and evidence, not conversational memory. Keep the controlling state small, current, and inspectable.
Operating commitments
- Retrieval over memory — externalize the plan, decisions, results, and handoffs; reread the relevant state before resuming.
- Evidence over reassurance — never call work verified, working, correct, or complete without a check run in the current execution context.
- Small tasks, whole-project alignment — finish bounded tasks while repeatedly checking the overall goal and acceptance criteria.
- Integration over accumulation — confirm every accepted task result appears consistently in the final deliverable.
- Convergence over first-fix stopping — after a correction, rerun the complete relevant check; stop only after a clean independent pass.
Route the work
Estimate scope before creating ceremony:
- Small task — do not use this skill.
- Large cross-domain project — use the workflow below.
- Large software project with GSD already installed — prefer GSD for milestone/slice/task state, worktrees, and autonomous software phases. Retain this skill's honesty and acceptance gates.
- Bounded software implementation with Superpowers already installed — use its design, plan, execution, and branch-completion workflow when it matches the harness.
Read references/software-backends.md before delegating. Never install or initialize an external controller merely because this skill triggered.
For any code change, apply references/coding-policy.md.
Execute the project
1. Define done
Write a concrete goal, deliverables, exclusions, constraints, and acceptance checks. Resolve only the ambiguities that materially change the project. Do not start producing final deliverables with an untestable definition of done.
2. Decompose
Create a small number of sequential stages such as research, design, build, integrate, verify, and polish. Break each stage into tasks with:
- A single outcome.
- Inputs and dependencies.
- A concrete output location.
- A check that proves completion.
- A status: pending, active, blocked, or verified.
3. Persist state
Use the project's existing issue tracker or planning convention when one exists. Otherwise create the minimal .project/ workspace described in references/workspace.md. Ask before creating project state when the user requested planning only or the target repository treats new root state as consequential.
Keep the master plan current. Record decisions and conventions once; do not make future sessions reconstruct them from chat.
4. Execute bounded tasks
For each ready task:
- Read the goal, current stage summary, dependencies, and relevant conventions.
- Mark only that task active.
- Perform the work without silently expanding scope.
- Run its specified check.
- Record output, evidence, remaining uncertainty, and any newly discovered work.
- Mark it verified only when its acceptance check passed.
- Update the stage summary and master plan before moving on.
Parallelize only independent tasks with non-overlapping outputs and a defined merge owner.
5. Integrate
When assembling task outputs, audit them one by one:
- Is each accepted result present?
- Are names, assumptions, formats, and versions consistent?
- Did later work invalidate an earlier task?
- Are generated artifacts traceable to their source task?
- Does the integrated result still meet the top-level exclusions and constraints?
6. Verify independently
Use the strongest applicable checks: executable tests, recomputation, lint/build, schemas, invariants, boundary cases, source/citation audit, visual inspection, or a genuinely independent review. Read references/verification.md.
After any failure, fix it and rerun the full relevant suite. Keep unchecked assumptions explicit.
7. Close or hand off
Complete only when:
- Every required task is verified or explicitly waived by the user.
- Every accepted task is incorporated.
- A fresh full pass finds no new issue.
- The final artifacts and evidence are discoverable.
- Remaining risks, deferred work, and reproduction instructions are recorded.
If pausing, leave a handoff containing current stage, last verified result, active blockers, exact next task, and commands or files needed to resume.
Communication
Report outcomes in this order:
- What changed or was produced.
- Checks run and their actual results.
- What remains unverified or blocked.
- The next ready task.
Do not expose the entire internal workspace unless requested. Never hide a mismatch to make progress look smoother.
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
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.

