execute
Skill0 starsChanged 2 months agoMITView on GitHub
The ATTENDED single-window execution mode: run a multi-phase SDD spec (produced by /substrate:architect-spec) with one implementing agent while a human co-pilots, pausing at each step/phase gate for approval. Invoke with a spec path (docs/tasks/ongoing/<feature>/<feature>-spec.md). Walks Phase N → Step N.M → Verify → Gate per docs/protocol/sdd/execution-format.md. This is the single-window (K=1, human-in-the-loop) alternative to the primary door /substrate:orchestrate, which runs K context-budget windows as an unattended parallel worktree fleet. Choose execute when you want to watch/adapt one window, or the spec fits one window and you prefer HIL. Best run in a fresh Claude session for a clean context window.
- Commits and pushes
57 other projects use a version of this.
Adopted by
Most recently taken up by
- LeonardiFranco/skillstheir own version, first seen Nov 8, 2026
- pfangueiro/claude-code-agentstheir own version, first seen Nov 8, 2026
- Yeachan-Heo/oh-my-claudecodetheir own version, first seen Nov 8, 2026
- crbnos/carbontheir own version, first seen Nov 7, 2026
- afaraha8403/balakittheir own version, first seen Nov 6, 2026
- chrislacey89/skillstheir own version, first seen Nov 4, 2026
Used with
Projects with it also use context-optimize, ralplan and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ibahgatSkillcontext-optimize
Improve context efficiency and cache stability for long or complex sessions.
Also in 3 projects that use this
ibahgatSkillralplan
Create a strict, quality-first execution plan with explicit gates and rollback-ready sequencing.
Also in 5 projects that use this
Yeachan-HeoSkillultragoal
Create and execute durable repo-native multi-goal plans over Codex goal mode artifacts.
Also in 4 projects that use this
buddyhSkilldeep-dive
Deep dive on any topic. Decomposes into subtopics, researches each in parallel with thorough source gathering, then synthesizes into a comprehensive report (4,000-10,000+ words, 40-100+ sources). Scales depth dynamically based on topic complexity. Use when asked for a deep dive, deep research, comprehensive analysis, or thorough investigation of a topic.
Also in 4 projects that use this
mukcoSkillprinciples
General software engineering principles — DRY, SOLID, YAGNI, KISS, naming, code smells, and refactoring. Language-agnostic. Use alongside language-specific skills.
Also in 3 projects that use this
team-attentionSkillcompound
This skill should be used when the user says "/compound", "compound this", "document learnings", "save what we learned", or after completing a PR. Extracts knowledge from PR context and saves to docs/learnings/.
Also in 3 projects that use this
- +1 more from ibahgat
Other versions57 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 57
- Yeachan-Heo/oh-my-claude-sisyphus★ 41ksame name
- Yeachan-Heo/oh-my-claudecode★ 41ksame name
- crbnos/carbon★ 2.8ksame name
- zaxbysauce/opencode-swarm★ 519same name
- ZaxbyHub/opencode-swarm★ 518same name
- Joonghyun-Lee-Frieren/oh-my-antigravity★ 214same name
- Joonghyun-Lee-Frieren/oh-my-gemini-cli★ 214same name
- team-attention/hoyeon★ 173same name
- Adrian333Dev/flow★ 39same name
- shipshitdev/skills★ 39same name
47 more not listed
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 repository14 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- add-doctrineSkill
- adoptSkill
- architect-specSkill
- deploySkill
- diagnoseSkill
- graph-specSkill
- initSkill
- migrateSkill
- orchestrateSkill
- quick-specSkill
- serve-beadSkill
- spoolSkill
- synthesize-sessionSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
23 rules. 0 are common, 15 are distinctive, and 8 are very specific to this project.
0
Common
Most of the 55 edited versions we read keep these.
15
Distinctive
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Run a spec phase-by-phase with gated verification, one window, with a human co-piloting.
- Very specific: line 50
If a tracker (tbd) is configured, read the DAG: bash docs/scripts/bead-graph.sh --epic <slug>.
- Very specific: line 51
No tracker, or no epic:<slug> beads → the spec has no partition to fan out.
- Distinctive: line 58
Offer the default, then honor the choice: If orchestrated would clearly win, surface it — "this spec partitions into a parallel fleet.
- Distinctive: line 73
Read the full spec at the provided path.
- Very specific: line 77
Every step has a ##### Verify block.
- Distinctive: line 80
If any structural check fails, stop and report which section is broken so the user can fix the spec before re-running.
- Distinctive: line 122
All commands exit 0 → step passes, continue to next step.
- Distinctive: line 127
If still failing, STOP and ask the user what to do.
- Distinctive: line 129
After the last step of the phase:
- Very specific: line 131
3e. Run phase gate: If a #### Gate block exists, run its commands via Bash.
- Very specific: line 145
Wait for explicit approval.
- Distinctive: line 151
Read the full integrated diff (this feature's changes vs.
- Distinctive: line 178
Do NOT push. The user decides when to push.
- Distinctive: line 197
MUST parse the spec before executing anything.
- Distinctive: line 198
MUST stop execution at any failed verify/gate — do NOT skip over failures.
- Distinctive: line 199
MUST pause for user approval between phases (after each gate).
- Very specific: line 200
MUST run the Doctrine Reconciliation phase — never skip the final phase — and apply any earned ratify-only doctrine change to …
- Distinctive: line 201
MUST commit after successful execution (step 6).
- Very specific: line 202
MUST NOT push to GitHub or deploy — that's for the user or /substrate:deploy.
- Very specific: line 203
MUST NOT invent steps beyond what the spec contains.
- Distinctive: line 204
SHOULD narrate each phase + step + verify result so the user sees liveness on long executions.
- Distinctive: line 205
SHOULD attempt one obvious fix when a verify fails (missing import, trivial type error).
“Common” is measured: kept by at least 6 in 10 of the 55 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/substrate:execute
Run a spec phase-by-phase with gated verification, one window, with a human co-piloting. This is
the attended execution mode — the single-window (K=1) case of substrate's context-budget
partition (agents-parallel-execution-doctrine.md §Grouping & windows), where one implementing agent
walks the whole spec and pauses at each gate for you.
Orchestrated is the default; attended is the deliberate choice. The primary execution door is
/substrate:orchestrate, which partitions the graphed DAG into K windows and runs them as an unattended parallel worktree fleet. Reach for attendedexecutewhen you want to watch and adapt one window as it runs, or the spec is small enough to be one window and you prefer human-in-the-loop over a fleet. Same partition, K collapsed to 1, human in the loop.
/substrate:architect-spec produces the spec; this skill executes it as the attended half of the SDD pipeline.
Arguments
<spec-path> — path to a spec at docs/tasks/ongoing/<feature>/<feature>-spec.md.
When to run
- A spec exists and follows the grammar in
docs/protocol/sdd/execution-format.md. - You want attended, human-in-the-loop execution of a single window — to watch/adapt as it runs,
or because the spec fits one window and a parallel fleet isn't worth the worktree overhead. For the
unattended parallel default, use
/substrate:orchestrateinstead. - You're in a FRESH Claude session (not the same session that drafted the spec). Clean context is the entire point.
When to REFUSE
| Signal | Redirect |
| Spec path missing or empty | Ask for the path. |
Spec does not contain a ## N. Prompt Execution Strategy section | Not executable — it's a design doc. Run /substrate:architect-spec to produce an executable version. |
| Same session already contains the brief + Q&A + architect outputs | Open a new terminal and re-invoke. A cluttered context window degrades step quality on long specs. |
Workflow
Step 0. Confirm attended is the right mode (orchestrated is the default)
You were invoked as the attended door, but orchestrated is substrate's primary strategy — so before walking phases, confirm this spec actually wants a single attended window rather than the parallel fleet. This is a routing check only; it never changes the attended steps below.
- Detect a graphed epic. Derive
<slug>from the spec directory (docs/tasks/ongoing/<slug>/). If a tracker (tbd) is configured, read the DAG:bash docs/scripts/bead-graph.sh --epic <slug>. No tracker, or noepic:<slug>beads → the spec has no partition to fan out; stay attended (Step 1 onward). - Assess whether orchestrated would clearly win. The parallel fleet earns its worktree overhead
only when all hold:
- a graphed
epic:<slug>exists, AND - some wave has ≥3 file-disjoint windows/beads, AND
- a tracker is configured.
- a graphed
- Offer the default, then honor the choice. If orchestrated would clearly win, surface it —
"this spec partitions into a parallel fleet;
/substrate:orchestrate <slug>is the primary door — run that instead, or stay attended here?" On explicit confirm to switch, hand off to/substrate:orchestrate <slug>(optionally--auto). Otherwise — no epic, fewer than 3 file-disjoint windows, no tracker, or the user prefers to stay attended — run the attended phase-by-phase path (Step 1 onward). Choosing attended is always valid: it's the deliberate single-window mode, not a fallback.
Never silently fan out. Spawning a parallel fleet is heavy and mutating (it cuts branches and
worktrees), so switching to orchestrated is always an explicit opt-in. Staying attended never spawns
a worktree. See /substrate:orchestrate and agents-parallel-execution-doctrine.md §Grouping &
windows for the fleet semantics.
Step 1. Load the spec
Read the full spec at the provided path. Verify it contains:
## N. Prompt Execution Strategysection (required perexecution-format.md).- At least one
### Phase N:subsection. - Every step has a
##### Verifyblock. - The final phase is "Doctrine Reconciliation" (mandatory, terminal per
spec-template.md§N).
If any structural check fails, stop and report which section is broken so the user can fix the spec before re-running.
Step 2. Parse the execution plan
Walk the spec's Prompt Execution Strategy section. For each phase, capture:
- Phase number + name
- Steps, each with: title, prompt content, verify commands, optional timeout
- Optional phase gate commands
Print a one-line summary per phase, then wait for approval before starting:
Execution Plan — <feature>
Phase 1: Domain Layer (3 steps, gate: pnpm app:compile && pnpm app:test)
Phase 2: Schema + Backend (4 steps, gate: pnpm app:compile && pnpm app:test)
Phase 3: Frontend (5 steps, gate: pnpm app:compile && pnpm app:test)
Phase 4: Integration + E2E (2 steps, gate: pnpm app:test:e2e)
Phase 5: Doctrine Reconciliation (1 step, gate: re-run epic gate on integrated tip)
Start Phase 1? (y / n)Step 3. Walk each phase
For each phase, in order:
For each step within the phase:
3a. Announce.
--- Phase <N>, Step <N.M>: <Title> ---3b. Execute. Carry out the step's prompt content using the appropriate tools (Write, Edit, Bash). Respect any "Tools to use" / "Tools to NOT use" hints in the step body.
3c. Run verify commands. Each ##### Verify list item is a shell command in backticks. Run them sequentially via Bash, in order. Apply the step's ##### Timeout if specified (default 180000ms).
3d. Interpret results.
- All commands exit 0 → step passes, continue to next step.
- Any command exits non-zero → step FAILS:
- Show the failing command and its output.
- Attempt one fix based on the error if it's obvious (missing import, type mismatch, missing dir).
- Re-run verify.
- If still failing, STOP and ask the user what to do.
After the last step of the phase:
3e. Run phase gate. If a #### Gate block exists, run its commands via Bash. Same semantics as verify.
3f. Pause for user approval. Print:
✔ Phase <N> complete.
Steps: <N> passed
Gate: <gate commands> → green
Summary: <one-paragraph of what changed in this phase>
Continue to Phase <N+1>? (y / n / pause)Wait for explicit approval. n or pause stops execution cleanly so the user can inspect before resuming.
Step 4. Final phase: Doctrine Reconciliation
The last phase is always Phase N: Doctrine Reconciliation per spec-template.md. It runs against the fully integrated feature and applies the doctrine change the code earned — it does not detect-and-queue. /substrate:execute is depth-0 and attended, so apply the mutation directly:
- Read the full integrated diff (this feature's changes vs. the pre-execution base) and the doctrines the spec loaded.
- Edit
docs/doctrine/**directly to codify what the landed code demonstrates — under the ratify-only rule: you may only relax an outdated rule, promote a pattern the code demonstrates, or add coverage the code exemplifies. You may never introduce a MUST / MUST-NOT the just-landed code violates. If reconciliation would require a stricter rule that invalidates shipped code, do not apply it here — leave it for/substrate:synthesize-session. - Re-run the gate on the integrated tip (
pnpm app:compile && pnpm app:lint && pnpm app:test). Green ⟹ the mutation was ratify-only and lands with the feature in the same commit. Red ⟹ not ratify-only; revert the doctrine edit and defer.
There is no doctrine-amendments.md, no type: doctrine-amendment bead, no handoff queue — behavior is identical regardless of bead-tracker. If no doctrine change is earned, this phase is a no-op.
Step 5. Archive the task
Once the Doctrine Reconciliation phase passes, archive the feature directory:
mkdir -p docs/tasks/completed
mv "docs/tasks/ongoing/<feature>" "docs/tasks/completed/<feature>"Step 6. Commit the execution
git add -A
git commit -m "feat(<feature>): execute spec
Spec: docs/tasks/completed/<feature>/<feature>-spec.md
Phases: <N>
Doctrine reconciled in-epic: <yes (docs/doctrine/… touched) | no change earned>
"Do NOT push. The user decides when to push.
Step 7. Handoff
✔ Spec execution complete.
Feature: <feature>
Phases completed: <N>
Archive: docs/tasks/completed/<feature>/
Doctrine reconciled in-epic: <docs/doctrine/… touched | no change earned>
Next:
- git push when you're ready
- Or continue iterating with /substrate:quick-spec, /substrate:architect-spec, or /substrate:deployConstraints
- MUST parse the spec before executing anything. A malformed spec must fail fast, not halfway through.
- MUST stop execution at any failed verify/gate — do NOT skip over failures.
- MUST pause for user approval between phases (after each gate). Non-negotiable — gated execution is the whole point of this skill.
- MUST run the Doctrine Reconciliation phase — never skip the final phase — and apply any earned ratify-only doctrine change to
docs/doctrine/**in-epic, then re-gate. Never queue it as adoctrine-amendmentor write adoctrine-amendments.md. - MUST commit after successful execution (step 6). One commit per spec = one revertable unit.
- MUST NOT push to GitHub or deploy — that's for the user or
/substrate:deploy. - MUST NOT invent steps beyond what the spec contains. The spec is binding during execution (per
_SPEC-STANDARD.md§3). - SHOULD narrate each phase + step + verify result so the user sees liveness on long executions.
- SHOULD attempt one obvious fix when a verify fails (missing import, trivial type error); escalate to the user if the first fix doesn't resolve.
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.

