Executing Plans

Announce on entry

I'm using the executing-plans skill as a fallback because subagent dispatch is unavailable in this harness. I'll execute the plan inline, task-by-task, with human-partner checkpoints. If any precondition fails, I will STOP rather than proceeding on best-effort.

Leyline works significantly better with subagent dispatch. If your harness has a subagent primitive, interrupt and switch to subagent-driven-development. The inline fallback inherits session bias that subagents are bias-free of by construction.

Hard gate

Do NOT run any task step, touch production code, or mark any task complete
until all preconditions are satisfied: (1) an approved plan exists at
docs/leyline/plans/<YYYY-MM-DD>-<feature-name>.md, (2) the current working
directory is inside the worktree recorded in the baseline note, (3) the
harness does NOT support subagent dispatch (if it does, use
subagent-driven-development instead), AND (4) for any task whose "Files:"
block touches a user-facing surface, the UX spec section covering that surface
is current and approved. If any check fails, STOP. Route (1) back to
writing-plans; route (2) back to using-git-worktrees; route (3) to
subagent-driven-development; route (4) back to design-brainstorming. This
applies to EVERY project regardless of perceived simplicity or obviousness.

Violating the letter of the rules is violating the spirit of the rules.

Core principle

An inline executor is biased by definition - the same session that helped write the plan is now executing it. The bias cannot be removed; it can only be compensated for with discipline: follow every step in the plan verbatim, run every verification, mark completion only on observable evidence, and ask the human partner before resolving any ambiguity. When in doubt, ask; do not guess.

Precondition check (STOP if not satisfied)

0. Resolve <feature-name> from the plan filename (same slug Stages 3 and 4 used).

1. Approved plan present. docs/leyline/plans/<YYYY-MM-DD>-<feature-name>.md exists. If missing, STOP and route to writing-plans.

2. Baseline readable; feature branch active. Same extraction as subagent-driven-development:

   baseline_path="docs/leyline/plans/<YYYY-MM-DD>-<feature-name>-baseline.md"
   recorded_path=$(grep -E '^- Worktree path:' "$baseline_path" | sed 's/^- Worktree path:[[:space:]]*//')
   recorded_branch=$(grep -E '^- Branch name:' "$baseline_path" | sed 's/^- Branch name:[[:space:]]*//')
   [ -n "$recorded_path" ] && [ -n "$recorded_branch" ] || { echo "malformed baseline note"; exit 1; }
   
   [ "$(git rev-parse --show-toplevel)" = "$recorded_path" ] || { echo "not in recorded worktree"; exit 1; }
   [ "$(git rev-parse --abbrev-ref HEAD)" = "$recorded_branch" ] || { echo "wrong branch"; exit 1; }
   

   If the worktree mismatches, cd into the recorded path. If the branch is wrong (for example main), STOP; do not commit any task on the wrong branch. Route back to using-git-worktrees.

3. Subagent primitive confirmed absent. executing-plans runs only when the current harness has no subagent primitive. If the primitive exists, STOP and switch to subagent-driven-development. Token pressure / rate limits / session length are NOT legitimate reasons to use this fallback.

4. Product spec approved (verbatim marker).

   grep -E '^Product spec approved - round [0-9]+ - [0-9]{4}-[0-9]{2}-[0-9]{2}$' "docs/leyline/specs/<YYYY-MM-DD>-<feature-name>-design.md"
   

   Missing marker: STOP and route to brainstorming.

5. Experience gate pre-execution (per-task). Before beginning any task whose "Files:" block touches a user-facing surface, verify all three conditions:

   • UX spec file exists: test -f docs/leyline/design/<YYYY-MM-DD>-<feature-name>-ux.md.

   • UX spec is approved: the file contains the verbatim approval marker:

     grep -E '^UX spec approved - round [0-9]+ - [0-9]{4}-[0-9]{2}-[0-9]{2}$' "<path>"
     

     Missing marker: STOP and route to design-brainstorming.

   • UX spec is current: last-modified newer than the plan, or approval round >= plan's referenced round. If not, route to writing-plans.

Process

digraph executing_plans_flow {
    Start [shape=doublecircle, label="Plan approved, inside worktree, no subagent primitive"];
    Load [shape=box, label="1. Load plan; skim end-to-end; raise concerns before starting"];
    Todo [shape=box, label="2. Create a TodoWrite entry per task"];
    Task [shape=box, label="3. Next task"];
    Gate [shape=diamond, label="Surface touched? Artifact current?"];
    LoopDesign [shape=box, label="STOP; loop to design-brainstorming"];
    Steps [shape=box, label="4. Execute each checkbox step in the task block verbatim"];
    Verify [shape=diamond, label="Verification matches Expected?"];
    Stop [shape=box, label="5. STOP - blocker, ask the human partner"];
    Commit [shape=box, label="6. Commit; mark task complete"];
    More [shape=diamond, label="More tasks?"];
    Final [shape=box, label="7. After final task, summarize and announce"];
    Successor [shape=box, label="Invoke requesting-code-review (+ requesting-design-review when surfaces touched)"];
    End [shape=doublecircle, label="Successor named"];

    Start -> Load -> Todo -> Task -> Gate;
    Gate -> LoopDesign [label="artifact missing"];
    Gate -> Steps [label="artifact current or non-UX task"];
    Steps -> Verify;
    Verify -> Stop [label="no"];
    Stop -> Task;
    Verify -> Commit [label="yes"];
    Commit -> More;
    More -> Task [label="yes"];
    More -> Final [label="no"];
    Final -> Successor -> End;
}

Checklist

Create one TodoWrite (or harness equivalent) entry per task. Within each task, follow these steps.

1. Load the plan. Read every task block end-to-end before beginning. Raise concerns with the human partner BEFORE touching code:

   • Any task that looks bigger than 5 minutes.
   • Any task without a failing-test step (except those with a declared non-code exception).
   • Any task whose "Files:" block references paths that do not exist.
   • Any task whose verification step does not name an expected output.

   Do not silently correct the plan. If the plan has a critical gap, STOP. Route back to writing-plans.

2. TodoWrite setup. Create one task entry per plan task. Mark entries in_progress as you enter each task; mark completed only when every step including verification is done.

3. Per-task execution. For each task:

   1. Experience gate. If the task touches a user-facing surface, confirm the UX artifact section covering the surface is current and approved. If not, STOP and route to design-brainstorming.
   2. Follow the steps exactly. Each checkbox is an action. Do not reorder, do not skip, do not improvise. For code tasks, the failing-test step is FIRST and must actually produce RED before the implementation step runs.
   3. Non-code exception. If the task declares the Exception line verbatim (Exception: <kind> task - no failing test. Verification: ...), follow the Exception's verification instead of a failing-test step.
   4. Run every verification. Paste the command output into the TodoWrite entry (or a local review log). If the output does not match the Expected line, STOP. Do not proceed to the next step on a mismatched verification.
   5. For UX tasks: trigger each state from the state matrix and record the observation. Run the accessibility verification procedure and paste its output. Do the RECONCILE step (side-by-side against the artifact).
   6. Commit. Follow the task's commit instruction (or the project's convention). Use a short but specific message.
   7. Mark complete. Only after verification passes and the commit lands.

4. Stop conditions. STOP and ask the human partner if any of these occur:

   • A blocker: missing dependency, failing test you cannot fix in a few minutes, unclear instruction.
   • A critical plan gap: the plan tells you to modify a file that does not exist, or name a function the spec does not describe.
   • You do not understand an instruction in the plan.
   • A verification fails repeatedly (2 or more attempts).
   • A finding surfaces that is outside the scope of the current task.

   Asking is cheap. Guessing is the dominant failure mode of inline execution.

5. Final summary. After the last task is marked complete, summarize what shipped in a structured message to the human partner:

   Feature: <feature-name>
   Branch: <branch name>
   Tasks: <count>
   Commits: <base-sha>..<head-sha>
   Surfaces touched: <yes / no>
   Test suite: <green / red-authorized / N/A>
   Stop conditions hit (if any): <list>
   

   Then transition.

Anti-patterns

• "I'll Skip To The Good Part Of The Task Block" - the steps are ordered for a reason. The failing-test step is first because it anchors everything after.
• "Silently Correct The Plan Mid-Execution" - the plan is the plan. If it is wrong, STOP and fix it in writing-plans with the human partner's approval. Do not patch as you go.
• "Guess When The Instruction Is Ambiguous" - ask. Guessing is indistinguishable from drift until a review catches it.
• "Skip The Verification Step When I'm Confident" - confidence is the rationalization that precedes every wrong commit. Run the verification.
• "Batch Commits Across Multiple Tasks For Speed" - the commit-per-task structure is what lets the stage-7 reviewer see task-level changes. Do not batch.
• "Run Verifications After All Tasks, Not During" - then every failure has many plausible causes. The per-task verification is what keeps causes local.
• "Use executing-plans Because Subagents Are Slow" - the fallback is for harnesses without the primitive, not for convenience. Use subagent-driven-development when the primitive exists.

Red flags

Forbidden phrases

Do not say:

• "Skipping verification; we know it works"
• "Batching these tasks for efficiency"
• "The plan is mostly right, I'll correct as I go"
• "This is a small deviation; it won't matter"
• "Running tasks inline because subagents are slow"

Output artifacts

• Per-task commits on the feature branch.
• TodoWrite entries marked complete as each task finishes.
• A short final-summary message to the human partner.

Successor

Invoking requesting-code-review (stage 7). All tasks executed, verifications passed. Dispatching the stage-7 reviewer for branch-level sign-off.

When any task touched a user-facing surface, additionally:

Invoking requesting-design-review (stage 7, parallel).

Missing-successor fallback

If any of the following is missing in this version of the plugin, STOP and name the missing piece:

• requesting-code-review - the Stage 7 branch-level reviewer.
• requesting-design-review - the Stage 7 branch-level design reviewer (required only when surfaces were touched).
• systematic-debugging - the route for unexpected failures hit during inline execution.

Do not improvise past any of them. Do not merge or ship without the stage-7 review.

Do not exit without naming and invoking the named successor.

Related

• ../../dev/stages/05-execute.md - canonical stage definition
• ../subagent-driven-development/SKILL.md - preferred successor when the harness supports subagent dispatch
• ../dispatching-parallel-agents/SKILL.md - for the rare case where 2+ independent problems surface during inline execution
• ../writing-plans/SKILL.md - produces the plan this skill consumes
• ../systematic-debugging/SKILL.md - the process to apply when a verification fails (planned Stage 6)