AGENT SKILL

insight-ooda-loop

Run a closed iterative research loop: Novel Insight Engine produces a novel falsifiable prediction → OODA turns it into an execution goal with Observe/ Orient/ Decide/ Act → finding feeds the next NIE round. Use when the user invokes /insight-ooda-loop or wants repeated cycles of insight generation + operational testing for hard research, debugging, theorem hunting, invention, or strategy. Hard stop at 20 rounds or solution_found.

License 许可
MIT
最近更新
2026-07-19
Stars
★ 1
Forks
1

01 前言 Front Matter

name
insight-ooda-loop
metadata
{ "short-description": "Closed-loop research engine: Novel Insight Engine → OODA execution → next context (max 20 rounds)", "canonical-codex-skill": "/Users/velocityworks/.codex/skills/insight-ooda-loop/SKILL.md" }
description
Run a closed iterative research loop: Novel Insight Engine produces a novel falsifiable prediction → OODA turns it into an execution goal with Observe/ Orient/ Decide/ Act → finding feeds the next NIE round. Use when the user invokes /insight-ooda-loop or wants repeated cycles of insight generation + operational testing for hard research, debugging, theorem hunting, invention, or strategy. Hard stop at 20 rounds or solution_found.
when-to-use
For sustained, closed-loop research or problem-solving where insight generation and falsification testing must alternate automatically. Trigger on 'run the insight ooda loop', 'insight-ooda', 'iterate insight and test', or any request for a multi-round engine that stops only on solution or round limit. Composes /novel-insight-engine + /ooda.
allowed-tools
[ "read_file", "grep", "list_dir", "search_replace", "write", "web_search", "web_fetch", "open_page", "open_page_with_find", "x_keyword_search", "x_semantic_search", "memory_search", "memory_get", "todo_write", "image_gen" ]
argument-hint
<overall research goal or problem context> [--rounds N (default 20)] [--stop-on-solution]

03 SKILL.md 正文

Insight OODA Loop — Grok Edition

Purpose

Run a closed research loop:

Novel Insight Engine -> prediction -> goal -> OODA -> finding -> next context

Use this for hard research, debugging, theorem hunting, invention, or strategy work where the user wants repeated insight generation and operational testing rather than one-off advice.

Required Composition

This skill composes two ported skills (once installed and discoverable via /skills or direct /novel-insight-engine and /ooda):

  • Novel Insight Engine (/novel-insight-engine): produces one novel, falsifiable insight with a prediction or decision rule (see its full contract).
  • OODA (/ooda): turns that prediction into action through Observe → Orient → Decide → Act with visible tokens (see its full contract).

When this skill runs, follow their core contracts. Do not paste their full text unless the user needs a reminder.

Loop Contract

Run exactly until one of these stop conditions is reached:

solution_found = true
round_number = 20   (or user-provided --rounds N)
user explicitly stops or redirects the loop

Do not stop because an intermediate artifact was created, a partial result was found, a useful reframing appeared, a commit landed, or the current round feels like a natural stopping point. Those are findings, not stop conditions.

If no solution is found, complete all rounds (up to the limit).

Each round has this shape:

Round N Context
  current problem context
  prior OODA finding, if any
  unresolved state

NIE
  one core insight
  falsifiable prediction or decision rule
  disconfirmation condition

Goal
  set or restate the OODA goal as the NIE prediction

OODA
  Observe: inspect evidence relevant to the prediction
  Orient: classify status relative to the goal
  Decide: choose the narrowest action that tests or advances the prediction
  Act: execute and verify

Finding
  supported, weakened, falsified, unresolved, or solution_found
  evidence
  next context for the following NIE round

Do not skip the prediction-to-goal handoff. Do not run OODA on the original vague problem if the NIE produced a sharper testable prediction.

Hard Execution Discipline

Before any tool call in a round, emit the round header and contract:

Round N / <limit>
NIE: <one insight and falsifiable prediction>
OODA Goal: <prediction converted into the round goal>
Disconfirmation: <what would weaken or falsify this prediction>
Planned Observe: <the exact evidence or artifact to inspect>

After the tool call or action, emit the finding block:

Finding: <supported/weakened/falsified/unresolved/solution_found>
Evidence: <concrete file, command, output, or observation>
Next Context: <input to the next NIE round>
Continue: <yes/no, with stop condition if no>

Do not compress multiple rounds into one explanation. Do not silently continue a round in the background. Do not decide on your own that a partial result is enough. Do not make exploratory tool calls before the current round has a visible NIE prediction, OODA goal, and disconfirmation condition.

State To Preserve

Maintain these fields in the working response or a durable artifact (use write to .insight-ooda-state.md or similar when the loop spans turns or sessions):

round_number
round_limit = 20 (or user value)
current_context
nie_core_insight
nie_prediction
nie_disconfirmation
ooda_goal
ooda_observe
ooda_orient
ooda_decide
ooda_act
finding_status
finding_evidence
next_context
solution_found
stop_reason

For code or research repositories, preserve durable artifacts when the loop creates evidence that future turns must inspect. Keep proof status, measured status, implementation status, and unresolved status separate.

Stop Rules

Stop immediately when one of these is true:

  • solution_found = true (the OODA evidence verifies the original user objective, not merely a smaller intermediate result)
  • round_number reaches the configured limit
  • user stops or redirects the loop

Blocked work is not a normal stop condition. If a round is blocked by missing information that cannot be discovered locally with the available tools, mark the round unresolved, state the exact missing information, and continue to the next round unless the user stops or redirects. Only stop for a blocker if the blocker makes every remaining round impossible.

When stopping, state:

stop_reason
rounds_completed
strongest_supported_finding
remaining_unresolved_point
next_best_action, if not solved

Do not use these as stop reasons:

  • tactical artifact landed
  • partial certificate found
  • interesting reframing found
  • commit completed
  • enough for now
  • natural checkpoint

Goal Tool Behavior

If the user explicitly asks to set a goal for the loop, use any available goal-tracking mechanism.

Inside the loop, treat each NIE prediction as the active OODA goal for that round. Do not create a new external goal every round unless the user explicitly asks for persistent goal tracking. A round-local OODA goal is enough.

If an existing persistent goal is active, keep the full goal intact and use the round prediction as a subgoal.

Output Shape (Active Run)

For an active run, use compact round blocks as shown in the Hard Execution Discipline section above.

Keep the output direct. The loop is for progress, not ceremony.

After each full round, consider writing a compact state snapshot with write so a future session or new context window can resume cleanly using read_file.

Guardrails

  • Do not turn the loop into brainstorming without action.
  • Do not run action (Act) without a falsifiable prediction from NIE.
  • Do not let a falsified prediction quietly become a weaker claim in the next round.
  • Do not add premises, fallback paths, generalized frameworks, or broad side quests unless the OODA evidence requires them for the current prediction.
  • Do not claim completion from indirect evidence.
  • For mathematical or scientific work (especially in this PGS workspace), never collapse measured evidence into proof language. Preserve the distinction between supported, unresolved, and invalidated states.
  • For factorization or PGS work, keep structure discovery separate from any factor_found style claims.
  • Always surface relevant PGS objects, invariants, DNI, GWR, etc. in the first Observe/Orient of any round that touches this domain (per AGENTS.md).
  • Respect the 20-round (or configured) hard cap. The engine is designed to be bounded.

Success Criteria

  • Every round produces the required visible header + finding blocks.
  • The loop terminates only on a true stop condition with a precise statement of outcome.
  • Progress (or explicit lack of progress) is documented such that a new session can resume without loss of context.
  • The final strongest_supported_finding is directly traceable to the original user objective via the chain of NIE predictions and OODA verifications.
  • No drift from the locked goal occurs across rounds.

Invocation Example

/insight-ooda-loop Close the toy-scale RSA-v2 PGS certificate using only deterministic structural invariants --rounds 12

The loop will generate the first NIE insight/prediction, convert it to an OODA goal, execute one disciplined cycle, feed the finding back, and continue until solution or limit.

This skill turns two powerful primitives (novel falsifiable insight + disciplined execution) into a persistent research engine.

04 同仓库更多 Skills

zfifteen
zfifteen/skills-catalog

Zotero

Use Zotero Desktop from Codex to enable/probe the local API, search a local Zotero library, list items/collections/tags, export BibTeX, insert citation keys into LaTeX or Markdown drafts, read indexed full text when requested, and import BibTeX/RIS records into Zotero through the connector server. Use when the user mentions Zotero, citations, references.bib, BibTeX export, local Zotero API, localhost:23119, or adding citations from a Zotero library.

★ 1
zfifteen
zfifteen/skills-catalog

zoom-apps-sdk

Use when using Apps SDK.

★ 1
zfifteen
zfifteen/skills-catalog

youtube

Obtain accurate, timestamped YouTube video transcripts (preferring manual captions over auto-generated) as the primary deliverable, then derive summaries, key claims, or notes from that transcript. Use when the user provides a YouTube URL (or video ID) and wants the transcript, a faithful summary, key claims with attribution, timestamped notes, or "what does this video actually say". The skill is transcript-first: never substitute metadata, title, description, or search snippets for the spoken content. Triggers: "youtube transcript", "get the transcript of this video", "summarize this youtube video", "what did they say in https://youtube.com/...", "key claims from this talk".

★ 1
zfifteen
zfifteen/skills-catalog

yeet

Publish local changes to GitHub by confirming scope, committing intentionally, pushing the branch, and opening a draft PR through the GitHub app from this plugin, with `gh` used only as a fallback where connector coverage is insufficient.

★ 1
zfifteen
zfifteen/skills-catalog

x-post-creator

Create X-ready post threads and co-located story packages from the active conversation or referenced project material. Use when the user says turn this into an X-post, /x-post, make this into an X thread, or wants plain-language social posts with supporting images saved to docs/x-posts/[story-name]/.

★ 1
zfifteen
zfifteen/skills-catalog

x-api

Search, read, and write X posts through the local X API MCP server (the same backend used by the Codex x-api skill). Use for searching X/x.com, reading threads and replies, creating lists, staging/publishing posts for @alltheputs, finding high-visibility reply targets for research outreach, ingesting public context, or any request that mentions /x-api or "search X".

★ 1