Meta Prompting

Meta-Prompting

Enhanced reasoning via /commands or natural language. Commands combine left-to-right: /verify /adversarial. Auto-trigger when context warrants -- note which pattern applied. Output: apply the pattern inline, then mark the result (e.g., VERIFIED ANSWER:, REVISED ANSWER:, confidence tier).

Patterns

`/think` | /show -- Show reasoning step-by-step: decision points, alternatives considered, why each accepted/rejected. With /think doubt: after each step, flag what could be wrong and why before proceeding.

`/adversarial` | /argue -- After answering, steelman the opposing case. 3 strongest counterarguments ranked by severity. Identify blind spots and unstated assumptions.

`/constrain` | /strict -- Tight constraints: 3 sentences max, cite sources, no hedging. Override inline: /constrain 5 sentences.

`/json` | /format -- Respond in valid JSON code block, no surrounding prose unless asked. Default schema:

{"analysis": "string", "confidence_score": 85, "methodology": "string", "limitations": ["string"]}

Custom keys: /json {keys: summary, risks, recommendation}

`/budget` | /deep -- Extended thinking space (~500 words) showing dead ends and reasoning pivots, then clearly separated final answer.

`/compare` | /vs -- Compare options as table. Default dimensions: speed, accuracy, cost, complexity, maintenance. Custom: /compare [dim1, dim2].

`/confidence` | /conf -- Rate each claim 0-100. Flag below 70 as SPECULATIVE. Group by tier: HIGH (85+), MEDIUM (70-84), LOW (<70). Include assumptions made and rate each 1-10 on confidence.

`/edge` | /break -- 5+ inputs/scenarios that break the approach. Code: null/empty, concurrency, overflow, encoding, auth bypass. Strategies: market conditions, timing, dependencies. Auto-triggers on: security, validation, parsing contexts.

`/verify-think` | /check -- Three phases: (1) Answer direct response, (2) Challenge 3 ways it could be wrong, (3) Verify investigate each, update if needed. Mark final as VERIFIED ANSWER: or REVISED ANSWER:. Distinct from the /ia-verify slash command, which runs the full pre-PR verification pipeline. Auto-triggers on: architecture decisions, critical choices, "Am I right?"

`/flip` | /alt -- Identify the approach you'd take by default and state it. Then propose an alternative that uses a different mechanism (different data structure, different layer, different abstraction). State the conditions under which the alternative beats the default. Override: /flip 3 for top 3 alternatives. Auto-triggers on: architecture decisions where the "easy" answer may break at scale.

`/assumptions` | /presume -- Before answering, list every implicit assumption in the question/task. Then answer with assumptions explicit. The assumption list is often more valuable than the answer. Auto-triggers on: architecture reviews, ambiguous requirements.

`/premortem` | /postmortem -- Assume the decision/project has already failed. Work backwards: what caused the failure? List 3-5 failure modes by likelihood. Focus on systemic risks, not edge cases.

`/tensions` | /perspectives -- Answer from two named opposing perspectives (e.g., security engineer vs. shipping PM). Focus output on where they disagree -- that's where the real insight lives. Override roles: /tensions [devops, security].

Combos

`/analyze` = /think + /edge + /verify-think -- Code reviews, architecture, security-sensitive work. Synthesize findings into a unified recommendation -- don't just concatenate pattern outputs. Auto-triggers on: code review requests.

`/trade` = /confidence + /adversarial + /edge -- Trade ideas, position analysis, market thesis. Auto-triggers on: trade/position discussions.

Conventions

• Separate combined pattern outputs with ---
• Keep core answer prominent -- patterns enhance, not bury the response
• Accept new pattern definitions mid-conversation ("Add /eli5 for explain like I'm 5") -- apply for the session

Verify

• Pattern marker present in output (e.g., VERIFIED ANSWER: for /verify)
• Core answer remains prominent -- meta-reasoning enhances, doesn't bury it

ia-meta-prompting Specification

Intent

ia-meta-prompting is a meta-class skill (patterns about prompts, agents, or skills themselves). Structured reasoning modifiers (/think, /verify, /adversarial, /edge, /confidence, /assumptions, etc.) to stress-test decisions, surface assumptions, or enumerate edge cases. Use when validating an important design, architecture decision, or ambiguous plan before committing.

Scope

In scope:

• Behaviors described in SKILL.md and routed via the should_trigger phrasings in distillery/tests/fixtures/triggers/ia-meta-prompting.jsonl.
• Updates to runtime behavior, structure, trigger precision, references, and validation.

Out of scope:

• Acting as the runtime instructions themselves (those live in SKILL.md).
• Trigger phrasings already covered by adjacent ia-* skills (validate-plugin flags >70% description overlap as DUPLICATE_TRIGGER).
• <!-- to fill in: domain-specific exclusions when the skill drifts -->

Trigger Context

• Class: meta
• Hook regex: plugins/whetstone/hooks/skill-patterns.sh -> SKILL_PATTERNS[ia-meta-prompting]
• Common requests (from fixture should_trigger):
• "argue against this architectural decision before we commit"
• "what could break if we deploy this change to production"
• "/think through whether this design holds under load"
• Should not trigger for (from fixture should_not_trigger):
• "add a new endpoint for user profile updates"
• "run the test suite and fix any failures"
• "write the migration script"

Source And Evidence Model

Authoritative sources:

• SKILL.md -- runtime instructions and reference routing.
• references/*.md -- bundled supplementary content (0 file(s)).
• distillery/tests/fixtures/triggers/ia-meta-prompting.jsonl -- positive and negative trigger phrasings under regression test.
• plugins/whetstone/hooks/skill-patterns.sh -- regex pattern that fires this skill.
• distillery/.eval-data/ia-meta-prompting/ -- harvested session examples (when present).

Data that must not be stored in this skill or its references:

• Secrets, credentials, tokens.
• Machine-specific filesystem paths (/home/..., /Users/..., ~/ai/...). The validator (MACHINE_PATH_LEAK) flags these as HIGH.
• Private URLs, customer data, or unredacted personal information.

Coverage matrix

Evaluation

Lightweight (run on every change):

python3 distillery/scripts/distiller.py validate-plugin --component ia-meta-prompting
python3 distillery/scripts/distiller.py test-triggers --skill ia-meta-prompting

Deeper (when behavior risk warrants):

python3 distillery/scripts/distiller.py dspy-eval ia-meta-prompting
python3 distillery/scripts/distiller.py diagnose-negatives ia-meta-prompting

Acceptance gates:

• validate-plugin --component ia-meta-prompting returns 0 HIGH findings.
• test-triggers --skill ia-meta-prompting returns F1 = 1.0 with floors of 5 should_trigger and 5 should_not_trigger.
• For dspy-eval, the composite score does not regress against the most recent saved baseline (see distillery/.eval-data/ia-meta-prompting/history.json).

Known Limitations

<!-- to fill in over time as drift surfaces. Default rule: any time diagnose-negatives surfaces a recurring failure pattern, document it here so future maintainers understand the trade-off the current implementation accepts. -->

Maintenance Notes

• Update SKILL.md when the runtime workflow, branch conditions, or output contract changes.
• Update this SPEC.md when intent, scope, evidence model, evaluation gates, or maintenance expectations change.
• Update the trigger fixture when adding new positive phrasings, removing stale ones, or expanding scope (the 5/5 floor is a hard validator gate).
• Update the hook regex in skill-patterns.sh whenever fixture positives expose a missed phrasing; verify F1 = 1.0 with eval-triggers before committing.
• Run the full release pipeline via /release -- never bump versions or update CHANGELOG.md from a per-skill edit.