agentleFS
qbicusqbicus/skills/skills/new-feature/SKILL.md

new-feature

Skill1 starsChanged 40 days agoMITView on GitHub

Drive a new feature through a mandatory staged workflow covering `nf-spec`, approval, `nf-design`, approval, `nf-tasks`, approval, and only then execution. Use when a user asks for `/new-feature`, wants disciplined feature delivery, or needs protection against skipping planning and starting implementation too early.

41 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use evidence-driven-testing, greploop-apps 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.
  • jaydubya818Skill

    evidence-driven-testing

    Capture observable runtime proof with video, screenshots, or measured output. Use when a change needs evidence beyond test reports.

    Also in 6 projects that use this

  • jaydubya818Skill

    greploop-apps

    Run greploop through @greptile-apps when the normal trigger rejects a large PR. Use only with the companion greploop skill.

    Also in 4 projects that use this

  • VercelSkill

    before-and-after

    Add existing screenshots or screen recordings to a GitHub pull request as a before/after or preview block. Use when a PR needs visual media attached to its description. Browser navigation and capture belong to agent-browser.

    Also in 6 projects that use this

  • BboyfaroukSkill

    greploop

    Iteratively improves a PR until Greptile gives it a 5/5 confidence score with zero unresolved comments. Triggers Greptile review, fixes all actionable comments, pushes, re-triggers review, and repeats. Use when the user wants to fully optimize a PR against Greptile's code review standards.

    Also in 6 projects that use this

  • u9401066Skill

    bug-fix

    Structured bug fixing workflow with root cause analysis and regression prevention. Triggers: BF, 修 bug, fix bug, debug, 除錯, 錯誤, error, 問題, issue, 修復, fix, 故障, fault, 異常, exception, 失敗, failure, 不能用, broken, 壞掉.

    Also in 3 projects that use this

  • SvampenSkill

    init-project

    Generates a tailored .claude/project-config.md for the current project. Asks about project type, quality check commands, documentation paths, insight tags, and review strictness.

    Also in 4 projects that use this

  • +1 more from jaydubya818

The file

Show its rules and text

We split this skill into the rules it gives an agent

61 rules. 0 are common, 25 are distinctive, and 36 are very specific to this project.

  • 0

    Common

    Most of the 39 edited versions we read keep these.

  • 25

    Distinctive

    This author's own choices. Worth a look.

  • 36

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Treat the request as a net-new delivery flow unless the user clearly says this is only a revision to an existing stage artifact.

    line 15
  2. Very specific:

    Use the same projectFolder and featureShortName for the decision log at <projectFolder>/_localnotes/<featureShortName>/decisions.md.

    line 16
  3. Very specific:

    If the feature includes Avalonia or cross-platform desktop UI, activate nf-avalonia as a supporting skill throughout spec, design, task …

    line 17
  4. Very specific:

    If the feature uses Avalonia, apply nf-avalonia guidance while drafting the spec.

    line 21
  5. Very specific:

    Write the spec to <projectFolder>/_localnotes/<featureShortName>/specs.md unless the user requests another filename.

    line 22
  6. Distinctive:

    Do not continue automatically after drafting.

    line 25
  7. Distinctive:

    Stop for spec approval

    line 27
  8. Distinctive:

    Ask for explicit approval of the spec.

    line 28
  9. Distinctive:

    If feedback is given, revise the spec and ask again.

    line 30
  10. Very specific:

    Do not begin nf-design until spec approval is explicit.

    line 31
  11. Very specific:

    After spec approval, log any material product or scope decisions using the decision skill.

    line 32
  12. Very specific:

    If the approved spec indicates Avalonia or cross-platform desktop UI, apply nf-avalonia guidance while drafting the design.

    line 36
  13. Very specific:

    Write the design to <projectFolder>/_localnotes/<featureShortName>/design.md unless the user requests another filename.

    line 37
  14. Distinctive:

    Keep the design traceable to the approved spec.

    line 38
  15. Very specific:

    Use the provider primary role for material design/architecture decisions.

    line 39
  16. Very specific:

    If the approved spec is substantial: true, run advisor in plan mode on the drafted design before presenting it for design approval.

    line 40
  17. Distinctive:

    Stop for design approval

    line 42
  18. Distinctive:

    Ask for explicit approval of the design.

    line 43
  19. Distinctive:

    If the user requests changes that only affect implementation approach, update the design and ask again.

    line 44
  20. Very specific:

    If the requested change alters approved requirements, scope, user-visible behavior, core entities, contracts, or validations, return to …

    line 45
  21. Distinctive:

    Do not create tasks and do not code until design approval is explicit.

    line 46
  22. Very specific:

    After design approval, log any material architecture, data-flow, interface, migration, security, or observability decisions using the …

    line 47
  23. Very specific:

    Run nf-tasks as the task stage

    line 49
  24. Very specific:

    Treat nf-tasks as the required tasks phase for this workflow.

    line 50
  25. Very specific:

    If the approved design includes Avalonia UI work, apply nf-avalonia guidance while generating UI-related task slices and dependencies.

    line 52
  26. Very specific:

    Write the task plan to <projectFolder>/_localnotes/<featureShortName>/dev-todos.md unless the user requests another filename.

    line 54
  27. Distinctive:

    Stop for task-plan approval

    line 56
  28. Distinctive:

    Ask for explicit approval of the task plan.

    line 57
  29. Distinctive:

    If the user asks to "just start coding" before this approval, refuse and point back to the required process.

    line 58
  30. Distinctive:

    If the requested change only affects sequencing, batching, rollout ordering, or task decomposition, revise the task plan and ask again.

    line 59
  31. Very specific:

    If the requested change alters implementation approach or architecture, return to nf-design, revise it, get design approval again, then …

    line 60
  32. Very specific:

    If the requested change alters requirements, scope, user-visible behavior, core entities, contracts, or validations, return to nf-spec, …

    line 61
  33. Very specific:

    After task-plan approval, log any material sequencing, rollout, or risk-management decisions using the decision skill.

    line 62
  34. Distinctive:

    Start implementation only after spec, design, and task-plan approvals are all explicit.

    line 65
  35. Very specific:

    Before execution, tell the user whether the approved task plan is sequential-only or parallel-eligible.

    line 66
  36. Very specific:

    If the plan is parallel-eligible, ask whether to execute in parallel or one task at a time.

    line 67
  37. Distinctive:

    If the user chooses sequential execution, implement in task order.

    line 68
  38. Very specific:

    If the user chooses parallel execution, hand batch orchestration to parallel-exec and run only approved parallel-safe batches concurrently …

    line 69
  39. Very specific:

    Update dev-todos.md during execution to mark tasks in-progress, blocked, or done.

    line 70
  40. Very specific:

    Treat execution as resumable: a later session should continue from the remaining approved tasks recorded in dev-todos.md.

    line 71
  41. Very specific:

    If the approved work includes Avalonia UI, keep nf-avalonia active during implementation.

    line 72
  42. Very specific:

    Use executor for bounded implementation/test loops, researcher for external evidence, fast only for low-risk mechanical routing, and …

    line 73
  43. Very specific:

    If substantially the same failure reaches the effective provider's behavior.advisor.repeatedFailureThreshold (resolve the merged provider …

    line 75
  44. Distinctive:

    After implementation, verify with the narrowest useful tests/build commands and report what was or was not run.

    line 76
  45. Very specific:

    If the approved spec is substantial: true, run advisor in complete mode after verification and before declaring the feature complete.

    line 77
  46. Very specific:

    If execution reveals a change that affects only task sequencing, return to nf-tasks approval.

    line 79
  47. Very specific:

    If execution reveals a change that affects implementation approach or architecture without changing approved requirements, return to …

    line 80
  48. Very specific:

    If execution reveals a change that affects requirements, scope, user-visible behavior, core entities, contracts, or validations, return to …

    line 81
  49. Very specific:

    Use the decision skill to append to <projectFolder>/_localnotes/<featureShortName>/decisions.md.

    line 85
  50. Distinctive:

    Do not log trivial mechanical edits, obvious refactors, or choices fully dictated by the approved artifacts.

    line 87
  51. Distinctive:

    When a previously approved direction changes, add a new decision entry explaining the change and note which prior decision it supersedes.

    line 88
  52. Distinctive:

    If a new implementation decision invalidates approved spec, design, or task assumptions, pause execution and return to the appropriate …

    line 89
  53. Distinctive:

    Do not merge stages into one response when approval is still pending.

    line 93
  54. Distinctive:

    Do not treat silence, impatience, or "looks fine" in another context as approval.

    line 94
  55. Distinctive:

    Do not let the user skip directly from idea to design, tasks, or code unless they explicitly instruct you to abandon this workflow.

    line 95
  56. Very specific:

    If the user wants to bypass the process, state that the new-feature skill is designed to prevent that shortcut and ask whether they want to …

    line 96
  57. Distinctive:

    If an upstream artifact changes after approval, return to the affected stage and re-run downstream stages that depend on it.

    line 97
  58. Very specific:

    Use the earliest invalidated stage rule: go back to nf-spec for requirement changes, nf-design for implementation-approach changes, and …

    line 98
  59. Very specific:

    When execution begins, state that all three approvals have been received and include the approved substantial: true|false classification.

    line 105
  60. Very specific:

    When parallel batches are active, show the current work split by agent or batch and reference parallel-exec if it is orchestrating the …

    line 108
  61. Distinctive:

    If execution is blocked because approval is missing, say so directly instead of partially implementing.

    line 109

“Common” is measured: kept by at least 6 in 10 of the 39 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

New Feature

Enforce a gated delivery process for new work. Do not skip stages, do not start coding until the task-plan approval gate is explicitly passed, and keep a running decision log for important choices made along the way.

Workflow

  1. Establish feature identity
  • Capture or confirm projectFolder and featureShortName.
  • Clarify the feature goal only if missing details block the spec.
  • Treat the request as a net-new delivery flow unless the user clearly says this is only a revision to an existing stage artifact.
  • Use the same projectFolder and featureShortName for the decision log at <projectFolder>/_localnotes/<featureShortName>/decisions.md.
  • If the feature includes Avalonia or cross-platform desktop UI, activate nf-avalonia as a supporting skill throughout spec, design, task planning, and execution.
  1. Run nf-spec
  • Produce the specification using the nf-spec skill.
  • If the feature uses Avalonia, apply nf-avalonia guidance while drafting the spec.
  • Write the spec to <projectFolder>/_localnotes/<featureShortName>/specs.md unless the user requests another filename.
  • Require the spec to record substantial: true|false under its execution classification. Treat normal feature work as substantial by default; only genuinely small/local work should be false.
  • Capture a Parallel Execution Preference in the spec: ask for preferred maximum parallel agents only when the work can plausibly be decomposed safely; otherwise record 1.
  • Do not continue automatically after drafting.
  1. Stop for spec approval
  • Ask for explicit approval of the spec.
  • Accept only clear approval language such as approved, approve spec, or an unambiguous revision request followed by final approval.
  • If feedback is given, revise the spec and ask again.
  • Do not begin nf-design until spec approval is explicit.
  • After spec approval, log any material product or scope decisions using the decision skill.
  1. Run nf-design
  • Produce the implementation design using the approved spec and the nf-design skill.
  • If the approved spec indicates Avalonia or cross-platform desktop UI, apply nf-avalonia guidance while drafting the design.
  • Write the design to <projectFolder>/_localnotes/<featureShortName>/design.md unless the user requests another filename.
  • Keep the design traceable to the approved spec; do not introduce hidden scope expansion.
  • Use the provider primary role for material design/architecture decisions. Routine repository exploration may use the executor role, and external documentation research may use the researcher role.
  • If the approved spec is substantial: true, run advisor in plan mode on the drafted design before presenting it for design approval. Resolve material concerns or surface them as NEEDS_DECISION before the approval gate.
  1. Stop for design approval
  • Ask for explicit approval of the design.
  • If the user requests changes that only affect implementation approach, update the design and ask again.
  • If the requested change alters approved requirements, scope, user-visible behavior, core entities, contracts, or validations, return to nf-spec, revise it, get spec approval again, then regenerate design and re-run downstream stages.
  • Do not create tasks and do not code until design approval is explicit.
  • After design approval, log any material architecture, data-flow, interface, migration, security, or observability decisions using the decision skill.
  1. Run nf-tasks as the task stage
  • Treat nf-tasks as the required tasks phase for this workflow.
  • Produce the task plan from the approved design.
  • If the approved design includes Avalonia UI work, apply nf-avalonia guidance while generating UI-related task slices and dependencies.
  • Require the task plan to mark sequential-only vs parallel-safe tasks, define any parallel batches, assign the intended execution role (primary|executor|researcher|fast) per task where delegation is useful, and assign Agent 1..N slots without exceeding the approved maximum from the spec.
  • Write the task plan to <projectFolder>/_localnotes/<featureShortName>/dev-todos.md unless the user requests another filename.
  1. Stop for task-plan approval
  • Ask for explicit approval of the task plan.
  • If the user asks to "just start coding" before this approval, refuse and point back to the required process.
  • If the requested change only affects sequencing, batching, rollout ordering, or task decomposition, revise the task plan and ask again.
  • If the requested change alters implementation approach or architecture, return to nf-design, revise it, get design approval again, then regenerate the task plan.
  • If the requested change alters requirements, scope, user-visible behavior, core entities, contracts, or validations, return to nf-spec, revise it, get spec approval again, then regenerate design and the task plan.
  • After task-plan approval, log any material sequencing, rollout, or risk-management decisions using the decision skill.
  1. Execute only after approval
  • Start implementation only after spec, design, and task-plan approvals are all explicit.
  • Before execution, tell the user whether the approved task plan is sequential-only or parallel-eligible.
  • If the plan is parallel-eligible, ask whether to execute in parallel or one task at a time. Do not exceed the approved maximum agent count from the spec.
  • If the user chooses sequential execution, implement in task order.
  • If the user chooses parallel execution, hand batch orchestration to parallel-exec and run only approved parallel-safe batches concurrently while preserving dependency order between batches.
  • Update dev-todos.md during execution to mark tasks in-progress, blocked, or done.
  • Treat execution as resumable: a later session should continue from the remaining approved tasks recorded in dev-todos.md.
  • If the approved work includes Avalonia UI, keep nf-avalonia active during implementation.
  • Use executor for bounded implementation/test loops, researcher for external evidence, fast only for low-risk mechanical routing, and escalate material choices to primary.
  • Executor scope is bounded: it may touch directly related files required for the assigned task but must report meaningful expansion and must not silently make architecture/contract/storage/rollout decisions.
  • If substantially the same failure reaches the effective provider's behavior.advisor.repeatedFailureThreshold (resolve the merged provider config; use the framework provider helper when available), run advisor in stuck mode before trying another materially similar approach.
  • After implementation, verify with the narrowest useful tests/build commands and report what was or was not run.
  • If the approved spec is substantial: true, run advisor in complete mode after verification and before declaring the feature complete. Resolve material concerns or report the blocking outcome.
  • During execution, append a decision entry whenever an implementation choice materially changes behavior, contracts, architecture, storage, rollout, operability, or testing strategy.
  • If execution reveals a change that affects only task sequencing, return to nf-tasks approval.
  • If execution reveals a change that affects implementation approach or architecture without changing approved requirements, return to nf-design approval and then re-run nf-tasks.
  • If execution reveals a change that affects requirements, scope, user-visible behavior, core entities, contracts, or validations, return to nf-spec approval and then re-run all downstream stages.

Decision Logging Rules

  • Use the decision skill to append to <projectFolder>/_localnotes/<featureShortName>/decisions.md.
  • Log decisions that future sessions would struggle to reconstruct from code alone.
  • Do not log trivial mechanical edits, obvious refactors, or choices fully dictated by the approved artifacts.
  • When a previously approved direction changes, add a new decision entry explaining the change and note which prior decision it supersedes.
  • If a new implementation decision invalidates approved spec, design, or task assumptions, pause execution and return to the appropriate approval gate.

Enforcement Rules

  • Do not merge stages into one response when approval is still pending.
  • Do not treat silence, impatience, or "looks fine" in another context as approval; get explicit confirmation tied to the current stage.
  • Do not let the user skip directly from idea to design, tasks, or code unless they explicitly instruct you to abandon this workflow.
  • If the user wants to bypass the process, state that the new-feature skill is designed to prevent that shortcut and ask whether they want to exit the skill-driven flow.
  • If an upstream artifact changes after approval, return to the affected stage and re-run downstream stages that depend on it.
  • Use the earliest invalidated stage rule: go back to nf-spec for requirement changes, nf-design for implementation-approach changes, and nf-tasks for execution-plan-only changes.

Output Rules

  • At each stage, state the current gate clearly: awaiting spec approval, awaiting design approval, or awaiting task-plan approval.
  • Reference the artifact path written for that stage.
  • Reference decisions.md whenever a new decision entry is added.
  • When execution begins, state that all three approvals have been received and include the approved substantial: true|false classification.
  • During execution, show task progress with completed, active, blocked, and remaining tasks or batches.
  • Include an explicit visual cue such as done: T001, T002 | working: T003 | blocked: T004 | remaining: T005, T006.
  • When parallel batches are active, show the current work split by agent or batch and reference parallel-exec if it is orchestrating the batch.
  • If execution is blocked because approval is missing, say so directly instead of partially implementing.

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.