Skip to content

xiduzo/wtf / skills/wtf.loop

wtf.loop

Verified

This skill should be used when a developer wants to autonomously execute all Traces under a fully-specified Epic or Feature — for example "go", "start building", "implement everything", "run the loop", "execute the feature", "build it all", "kick it off". Requires that the Epic/Feature/Trace tree is fully written before starting. Chains implement → verify → PR → re-aim for every Trace — spine-first within each Feature, parallel across Features — with targeted human-in-the-loop gates for contradictions, ambiguities, and every change to a Trace Plan's scenario set.

evals/evals.json

View on GitHub
{
  "skill_name": "wtf.loop",
  "evals": [
    {
      "id": 1,
      "prompt": "Run wtf.loop for Feature #1 — Settlement status. Test mode: skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use the graph from `evals/fixtures/linear-chain.json` as the pre-built input — its `feature`, `nodes`, and `edges` fields replace the output of `gh issue view` + `gh sub-issue list` + `gh issue-dependency list`. Assume staged delivery. Skip AskUserQuestion in step 3 and auto-approve the execution plan. Output only the proposed execution plan (step 3 format).",
      "expected_output": "An execution plan with one Feature unit whose Trace sequence runs strictly sequentially: Skeleton #10 first and alone, then #11, then #12. #11 and #12 are serialized because their `## Impacted Areas` overlap on src/settlements/status/, not because of dependency edges (none exist).",
      "files": [
        "evals/fixtures/linear-chain.json"
      ],
      "expectations": [
        "Trace #10 (the Skeleton) is first in the Trace sequence and runs alone",
        "Trace #11 appears after #10 and before #12",
        "Trace #12 is last in the Trace sequence",
        "No two of traces #10, #11, #12 are presented as parallel",
        "The plan attributes the serialization of #11 and #12 to shared files (src/settlements/ or StatusService), not to dependency edges"
      ]
    },
    {
      "id": 2,
      "prompt": "Run wtf.loop for an Epic containing Features #20, #21, #22. Test mode: skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use the graph from `evals/fixtures/parallel-features.json` as the pre-built input — its `nodes` array and `edges` object replace the output of `gh sub-issue list` + `gh issue-dependency list`. Assume staged delivery. Skip AskUserQuestion in step 3 and auto-approve the execution plan. Output only the proposed execution plan (step 3 format).",
      "expected_output": "An execution plan with 1 phase and 1 sub-phase listing all 3 Features (#20, #21, #22) as parallel units with no ordering constraints between them. Each Feature's single Trace (#23, #24, #25) runs inside its own Feature unit.",
      "files": [
        "evals/fixtures/parallel-features.json"
      ],
      "expectations": [
        "The plan shows 1 phase",
        "Features #20, #21, and #22 are all listed in the same sub-phase",
        "The sub-phase is marked as parallel across Features",
        "No Feature is described as blocked by another Feature in the plan",
        "Traces #23, #24, and #25 each appear inside their own Feature unit, not as free-standing parallel units"
      ]
    },
    {
      "id": 3,
      "prompt": "Run wtf.loop for an Epic containing Features #30, #31, #32, #33. Test mode: skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use the graph from `evals/fixtures/diamond-file-conflict.json` as the pre-built input — its `nodes` array and `edges` object replace the output of `gh sub-issue list` + `gh issue-dependency list`. Assume staged delivery. Skip AskUserQuestion in step 3 and auto-approve the execution plan. Output only the proposed execution plan (step 3 format).",
      "expected_output": "An execution plan where Phase 1 runs Feature #30 alone; Phase 2 splits Features #31 and #32 into separate sub-phases due to the shared src/api/ file area conflict; Phase 3 runs Feature #33 after both #31 and #32 complete.",
      "files": [
        "evals/fixtures/diamond-file-conflict.json"
      ],
      "expectations": [
        "Phase 1 contains only Feature #30",
        "Features #31 and #32 are not in the same sub-phase",
        "Feature #33 appears in a phase after both #31 and #32",
        "The plan shows at least 3 phases",
        "The plan notes a file conflict or file area overlap between Features #31 and #32 involving src/api/"
      ]
    },
    {
      "id": 4,
      "prompt": "Run wtf.loop in test mode. Skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use the spec from `evals/fixtures/contradiction-spec.json` — its `feature` and `traces` fields replace the output of `gh issue view` and `gh sub-issue list`. Skip pre-flight checks 2a, 2c, and 2d. Run only step 2b (contradiction scan). Output only the findings from step 2b.",
      "expected_output": "Step 2b surfaces an actor contradiction: the Feature declares 'Merchant' as the sole actor and its canonical scenarios use Merchant, but Trace #10's Scenario Claim and synced Gherkin copy use 'Admin user'. The claimed scenario names do not exist in the Feature's canonical scenario set. The finding identifies both the Feature and Trace #10 as the conflicting sources and does not proceed to step 3.",
      "files": [
        "evals/fixtures/contradiction-spec.json"
      ],
      "expectations": [
        "The contradiction scan identifies a mismatch between the Feature actor ('Merchant') and Trace #10's claimed/synced Gherkin actor ('Admin user')",
        "The scan notes that Trace #10's claimed scenario names do not match the Feature's canonical scenarios",
        "The finding names both the Feature (#62) and Trace #10 as the conflicting sources",
        "The skill does not auto-resolve the contradiction or proceed to proposing an execution plan",
        "The output is framed as a pre-flight gate requiring human resolution"
      ]
    },
    {
      "id": 5,
      "prompt": "Run wtf.loop in test mode. Skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use the graph from `evals/fixtures/external-blocker.json` — its `nodes` array and `edges` object replace the output of `gh sub-issue list` and `gh issue-dependency list`. Skip pre-flight checks 2a, 2b, and 2c. Run only step 2d (dependency validation). Output only the dependency validation findings.",
      "expected_output": "Step 2d correctly classifies issue #999 (present in Feature #40's blocked_by list but absent from the node list) as an external blocker. It surfaces this as a pre-flight gate: the loop cannot proceed until #999 is merged or closed. It does not proceed to propose an execution plan.",
      "files": [
        "evals/fixtures/external-blocker.json"
      ],
      "expectations": [
        "Issue #999 is classified as an external blocker (not in the current run's node list)",
        "The external blocker is surfaced as a pre-flight gate rather than silently ignored",
        "The skill does not proceed to propose an execution plan (step 3) while #999 is unresolved",
        "The dependency chain is correctly described: #999 blocks Feature #40, which blocks Feature #41"
      ]
    },
    {
      "id": 6,
      "prompt": "Run wtf.loop for Feature #5 — Payment settlement. Test mode: skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use the graph from `evals/fixtures/sibling-traces-parallel.json` as the pre-built input — its `feature`, `nodes` (each with its Trace body), `edges`, `pull_requests`, and `remote_branches` fields replace the output of `gh issue view`, `gh sub-issue list`, `gh issue-dependency list`, `gh pr list`, and `git ls-remote`. Delivery is `staged` per the fixture's `config`. Skeleton #10 is already merged and its branch is gone. Skip AskUserQuestion in step 3 and auto-approve the execution plan. Output only the proposed execution plan (step 3 format), including for every unfinished Trace the branch it is cut from and its PR base.",
      "expected_output": "An execution plan with one Feature unit. Skeleton #10 is shown as merged and is not scheduled. Extension #11 and Deepening #12 sit in the same trace sub-phase and build in parallel (disjoint Impacted Areas), both cut from feature/5-payment-settlement. #11 joins first and targets feature/5-payment-settlement. #12 joins after it: it rebases onto trace/11-… and targets it, so the Feature keeps one linear stack. Deepening #13 sits in the next trace sub-phase because it shares src/settlements/export/ with #11. It is cut from and targets the stack tip trace/12-… once #12 has joined, without waiting for any Trace PR to merge.",
      "files": [
        "evals/fixtures/sibling-traces-parallel.json"
      ],
      "expectations": [
        "Traces #11 and #12 are placed in the same trace sub-phase and marked as building in parallel",
        "Trace #13 is placed in a trace sub-phase after #11 (never alongside it)",
        "Trace #11 is cut from and targets the feature branch feature/5-payment-settlement, because Skeleton #10 already merged",
        "Trace #12 joins the stack after #11: it rebases onto trace/11-… before its PR opens and its PR base is trace/11-…, so #11 and #12 never both target feature/5-payment-settlement",
        "Trace #13 is cut from and targets trace/12-… (the stack tip once #12 has joined), and the plan states it does not wait for any Trace PR to merge",
        "The plan shows Feature #5 as one linear stack (feature/5-… ← #11 ← #12 ← #13), not a tree",
        "Skeleton #10 is shown as already merged and is not scheduled for implementation again"
      ]
    },
    {
      "id": 7,
      "prompt": "Run wtf.loop for Feature #7 — Settlement notifications. Test mode: skip step 0 (gh setup already confirmed) and step 1 (hierarchy already fetched). Use `evals/fixtures/trunk-delivery-close.json` as the pre-built input — its `config` field is the contents of `.wtf/config.json`, and its `feature`, `nodes`, `edges`, `pull_requests`, and `remote_branches` fields replace the output of `gh issue view`, `gh sub-issue list`, `gh issue-dependency list`, `gh pr list`, and `git ls-remote`. Skip AskUserQuestion in step 3 and auto-approve the execution plan. Do not run git or gh. Output: the proposed execution plan (step 3 format); then, for each Trace, the branch it is cut from, its PR base, and the closure lines its PR body will carry; then the step 5 Feature completion action for this unit.",
      "expected_output": "A plan headed `delivery: trunk`. No feature branch is created or targeted. Skeleton #30 is cut from main and its PR targets main. Deepening #31 stacks on trace/30-… (PR base trace/30-… while #30 is open, retargeted to main by GitHub when #30 merges with its branch deleted). #31 is the last plan entry, so its PR body carries `Closes #31` and `Closes #7` on separate lines. Step 5 opens no feature PR; it verifies that the Feature issue closed via the final Trace PR.",
      "files": [
        "evals/fixtures/trunk-delivery-close.json"
      ],
      "expectations": [
        "The plan header or delivery line shows `trunk`",
        "No feature/7-… branch is created, cut from, or used as a PR base anywhere in the output",
        "Trace #30's PR base is main; Trace #31's PR base is either the trace/30-… branch (stacked) or main — never a feature branch",
        "The final Trace PR (#31) body carries `Closes #7` on its own line in addition to `Closes #31`",
        "Step 5 does not open a feature PR; it checks that Feature #7 closed via the final Trace PR"
      ]
    }
  ]
}
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall