Skip to content

rse/ase / plugin/skills/ase-task-grill

ase-task-grill

Verified

Interview the user relentlessly about the task plan until reaching a shared understanding, resolving each branch of the question decision tree. Use when the user wants to stress-test a plan, get grilled on their plan, or mentions "grill me" or "grill plan".

@${CLAUDE_SKILL_DIR}/../../meta/ase-control.md @${CLAUDE_SKILL_DIR}/../../meta/ase-skill.md @${CLAUDE_SKILL_DIR}/../../meta/ase-dialog.md @${CLAUDE_SKILL_DIR}/../../meta/ase-getopt.md

$ARGUMENTS

@${CLAUDE_SKILL_DIR}/../../meta/ase-format-task.md @${CLAUDE_SKILL_DIR}/../../meta/ase-common-task.md @${CLAUDE_SKILL_DIR}/../../meta/ase-common-ground.md @${CLAUDE_SKILL_DIR}/../../meta/ase-common-grill.md

Procedure

  1. Determine Task:

    1. Set initially, with any leading and trailing whitespace stripped. Inherit the always existing from the current context. Inherit the always existing from the current context. Do not output anything.

    2. If is not a positive integer, only output the following and then immediately STOP processing the entire current skill:

    3. Resolve the plan sections under focus:

      Each section in selects its focus areas of the grilling: SPECIFICATION selects DOMAIN and INTERFACE, DESIGN selects ARCHITECTURE and IMPLEMENTATION, and VERIFICATION selects REGRESSION and CONFIRMATION. Do not output anything.

      As the severity follows from the focus area, the focused sections yield at most the severity MUST (SPECIFICATION) resp. SHOULD (DESIGN, VERIFICATION). Hence, step 3.2 below caps the --until severity at the highest severity the remaining sections can yield, so a grilling of only DESIGN or VERIFICATION does not stop before asking anything.

    4. React on task id:

  2. Determine Task Plan:

    1. Determine the current task plan content:

    2. Determine the grilling scope of each section in , in order to grill an already grilled section again only where questions are still open:

      • If is true, ignore any grilled:<section/> tag: the scope of every section is the entire section, so already grilled sections are re-grilled completely. Skip the following two cases.

      • If the Tags: frontmatter key of carries the tag grilled:<section/> (with being the lower-case name of the section) and the section contains no bullet-point in checkbox state [?], the section is already grilled and nothing is open: remove it from and only output the following :

      • If the Tags: frontmatter key carries the tag grilled:<section/> and the section contains at least one bullet-point in checkbox state [?], restrict the scope of the section to exactly these [?] bullet-points, so the grilling re-asks only the still open questions.

      • Otherwise, the scope of the section is the entire section.

      Independent of the scope, bullet-points in checkbox state [-] (cancelled) or [>] (deferred) are inert: they are never subject of a question and their checkbox is never changed.

  3. Iterate Over Task Plan Aspects:

    1. Understand what "grilling" is about:

    2. Set to , but lowered to SHOULD if it is MUST and does not contain SPECIFICATION, as only SPECIFICATION yields questions of severity MUST. Do not output anything.

    3. Perform at most grilling rounds, numbered (1-) -- the round count is a maximum only, as every round can stop the grilling early in its item 2 below.

      For each round:

      1. INITIALIZE ROUND:

        Explicitly start from scratch from only the current -- as updated by all previous rounds -- and forget all questions and answers gathered in previous rounds -- except the grounding facts . Set to GRILLING ROUND <m/>/<getopt-option-rounds/> if is greater than 1, or to GRILLING otherwise (a single round needs no round numbering). Do not output anything.

      2. DETERMINE QUESTIONS:

        Focus only on the Focus Areas selected by -- the themes of the focused sections, within the grilling scope of each section as determined in step 2.3, while the entire stays the context -- and check the mentioned Indicators. For a section restricted to its [?] bullet-points, derive the questions from exactly these bullet-points only. Never raise a question about a [-] or [>] bullet-point. Create a decisions/questions tree for the questions, capturing the dependencies between the decisions.

        For each question, set <items-N/> to the bullet-points of
        <task-content/> the question is about -- possibly *none*,
        if the question concerns an aspect the plan does not
        cover yet.
        

        If is true, skip the items 3, 4, and 6 of this round, perform item 5 of this round (as the focused sections are clear, they are tagged as grilled), then skip all remaining rounds and continue with step 4.

      3. INTERACTIVE DIALOG:

        Treat as the combined free-text answers to all questions of the round and continue with item 4 below. In the following, you MUST NOT use your built-in tool! Instead, you MUST just show a custom dialog according to the expanded custom-dialog definition. You MUST closely follow this definition. The dialog below carries the fixed answer option SKIP GRILLING, dispatched as follows:

        • If a is SKIP GRILLING, ask no further questions -- all questions of this round stay unanswered --, continue with items 4 to 6 below (updating and saving the plan with the answers of the previous rounds), and afterwards skip all remaining rounds and continue with step 4.

        • If a is CANCEL, only output the following and then immediately STOP processing the entire current skill, not updating the plan with the current round (the plan updates saved by previous rounds are kept):

        Show a custom dialog. Its only answer option is the fixed one, so the user normally answers all aspects in one free-text reply:

        Dispatch SKIP GRILLING and CANCEL as defined above. Otherwise, strip any leading OTHER: prefix from and treat the remainder as the combined free-text answers to all questions of the round.

      4. MERGE ANSWERS INTO PLAN:

        Merge all gathered answers in of the round -- the combined reply -- exclusively back into .

        A question not addressed by the combined reply accepts its answer marked with ★ (the recommendation) if one exists, and otherwise stays unanswered. Additionally, record the open questions in the checkboxes of the body bullet-points, changing only checkboxes in state [ ] or [?] and leaving every [/], [x], [-], and [>] checkbox untouched:

        • For each answered question, set the checkbox of every bullet-point in to [ ], as the question is resolved now.

        • For each unanswered question, set the checkbox of every bullet-point in to [?], as the question is still open. If is empty, add a new [?] bullet-point to the section of the question's focus area , with the corresponding , a

        Set to all bullet-points of which were changed or added in this round. Do not output anything, except for the following warning:

      5. SAVE PLAN:

        For each section in , set the tag grilled:<section/> -- with being the lower-case name of the section (specification, design, or verification) -- in the Tags: frontmatter key of : keep an already present identical tag as is, otherwise append it, keeping all other already present tags (including grilled: tags of other sections) and creating the whole key at its position in the key order of the plan if the plan carries none. A fully grilled plan hence carries the tags grilled:specification, grilled:design, grilled:verification.

      6. SHOW PLAN CHANGES:

  4. Show Plan Diff:

    If is empty, set (none). Otherwise, set to , emitted verbatim, inside a fenced code block with the info string diff, whose fence is one backtick longer than the longest backtick run within (and at least three backticks long).

    Then only output the following , which shows the cumulative change of all rounds -- independent of and of , whose exposure rules are explicitly overridden here, as the user explicitly requested the diff:

  5. Decide Next Step:

    1. Determine next step:

      Next Step: How would you like to proceed with the plan? DONE: Stop processing. GRILL: Re-grill the plan again. EDIT: Hand off plan to editing. DRAFT: Hand off plan to implementation drafting. IMPLEMENT: Hand off plan to implementation.

    2. Check the tool and dispatch accordingly:

      • If is DONE or CANCEL: Only output the following and then STOP, without output of any further information.

      • If is GRILL: Set --force --rounds --until --focus . Set --perspective , with the value shell-quoted if it contains spaces or special characters. Set --ground. Set --auto. Set --quiet. Set --diff. Set --int-interactive. Only output the following and then call the tool Skill(skill: "ase:ase-task-grill", args: "<args/>") to invoke the ase:ase-task-grill skill again in order to re-grill the updated plan with the same grilling options. Immediately stop processing the current skill once the Skill tool was used.

      • If is EDIT: Set --quiet. Set --interactive. Set --focus . Set --ground. Only output the following and then call the tool Skill(skill: "ase:ase-task-edit", args: "<args/>") to invoke the ase:ase-task-edit skill in order to edit the updated plan. Immediately stop processing the current skill once the Skill tool was used.

      • If is DRAFT:

      • If is IMPLEMENT:

finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 319kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 319kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 319kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 196kUpdated Claude CodeCodexVerifiedInstall