research-continuity
Preserve durable research state, detect drift in reasoning or framing, and produce artifacts that let a future session (or different agent) continue high-stakes work without loss of context or accidental revival of invalidated ideas. Use when the user says "preserve continuity", "capture project memory", "shape feels wrong", "what should the next session do first", "hand off this research", or when ending a long thread and wanting explicit state. Works especially well with memory tools, todo_write, and the research-meeting skill.
- License 许可
- MIT
- 最近更新
- 2026-07-19
- Stars
- ★ 1
- Forks
- 1
01 前言 Front Matter
- name
- research-continuity
- metadata
- { "short-description": "Durable research memory, drift detection, and clean handoff artifacts" }
- description
- Preserve durable research state, detect drift in reasoning or framing, and produce artifacts that let a future session (or different agent) continue high-stakes work without loss of context or accidental revival of invalidated ideas. Use when the user says "preserve continuity", "capture project memory", "shape feels wrong", "what should the next session do first", "hand off this research", or when ending a long thread and wanting explicit state. Works especially well with memory tools, todo_write, and the research-meeting skill.
- when-to-use
- Any time continuity, durable memory, or early warning about conceptual drift is part of the request. Strong for multi-session research programs, handoffs between agents, or before compaction / context loss.
- allowed-tools
- [ "read_file", "grep", "list_dir", "write", "search_replace", "memory_search", "memory_get", "todo_write", "web_search", "web_fetch", "open_page", "x_keyword_search", "x_semantic_search" ]
- argument-hint
- <optional: specific thread, hypothesis, or 'full project state'>
03 SKILL.md 正文
Research Continuity
Purpose
Ensure that the critical distinctions, proved results, invalidated paths, and next actions survive across sessions, agents, compactions, or handoffs. The skill forces explicit capture of state in a form that is machine- and human-readable and that actively resists common failure modes (drift, revival of dead ideas, loss of epistemic status).
This is a direct Grok port and enhancement of the Codex "research-continuity" skill, deeply integrated with Grok's memory_* tools, todo_write, and the workspace's AGENTS.md / PROOF.md contracts.
When the Skill Is Active
The following distinctions must be made explicit in every continuity artifact:
- hypothesis
- measured result
- audit result
- proof result (or structural certificate)
- unresolved state
- invalidated rule / law / selector
Required Behavior on Invocation
-
Identify the Minimal Durable Set
-
Write Durable Artifacts
- Prefer repo-local files under
research/continuity/,docs/, or a slug-named handoff directory. - Use
writeorsearch_replacewith LF endings. - Also write or update relevant memory entries via the memory tools when appropriate for cross-project knowledge.
- Produce or update:
- A
STATE.mdorHANDOFF.mdanswering the five questions below. - Explicit test or reproduction commands.
- Updates to TODO lists or
todo_writeentries.
- A
- Prefer repo-local files under
-
Early Drift Detection (Mandatory Language)
- If the work's shape is starting to drift, say so immediately and plainly:
- "Shape feels wrong: ..."
- "This is starting to look like progress theater."
- "This is unresolved; the wording should not imply solved."
- "The code is using a classical gate before the named PGS rule."
- "PGS-first frame violated — classical selector used as inference engine."
- Name the concrete drift and the exact corrective action.
- If the work's shape is starting to drift, say so immediately and plainly:
-
PGS / Prime-Gap-Structure Contract
- In this workspace (or any PGS research): the continuity artifact must begin with the current ordered prime-gap state, divisor-count field, DNI status, GWR application, endpoint chains, chamber resets, etc.
- Never allow a continuity artifact to treat a proved PGS law as probabilistic or "empirically supported so far."
-
Second-Opinion / Pressure Integration
- For high-stakes decisions, explicitly recommend or invoke the
second-opinionskill (or equivalent viatasksubagent or codex-bus) with:- The exact question
- Relevant code excerpts / outputs / stats
- Known failed rules
- Current proposed plan
- Ask for critique, hidden assumptions, falsification paths, and one concrete next action. Follow up on material uncertainty.
- For high-stakes decisions, explicitly recommend or invoke the
Output Shape (The Five Questions)
A good continuity artifact always answers, in order:
- What is the current strongest supported claim? (With locator to proof / certificate / benchmark.)
- What is explicitly not proved or not solved? (Unresolved state, open chambers, missing invariants.)
- What failed and must not be revived? (Invalidated selectors, rules, assumptions, with dates and reasons.)
- What should the next session (or next agent) do first? (Smallest high-leverage action with exact command or file to edit.)
- What command or test reproduces the current state? (Reproducible one-liner or tool sequence.)
Success Criteria
- A new session or different agent can read the artifact(s) and pick up the work at the correct epistemic level without re-deriving or accidentally reviving dead paths.
- Drift warnings are issued early enough to be actionable.
- All five questions have crisp, locatable answers.
- In PGS work: the artifact itself demonstrates PGS-first reasoning.
Guardrails
- Avoid broad frameworks or ceremonial process documents. The artifact must be lean and actionable.
- Use LF line endings.
- Prefer writing state into existing contracts (AGENTS.md updates, PROOF.md addenda, spec/ files) over new ceremonial files.
- This skill does not execute the next action; it only captures the state for later execution.
- Higher-priority instructions (safety, explicit user contracts) always win.
Grok Environment Integration
memory_search+memory_getare first-class tools for both reading prior continuity state and writing new durable knowledge.todo_writeis the natural companion for tracking the "what next session should do" items.write/search_replacecreate the on-disk handoff files.- Synergy with
research-meeting(structured deep dives that produce minutes to be captured here) andcodex-bus(when the handoff must be persisted across truly separate agent instances via the ledger). oodaoractualizecan be used after a continuity artifact is written if immediate action is also required.
Example Minimal Artifact (for a PGS thread)
# Continuity Handoff — Reciprocal Transport Closure (2026-05-24)
**Strongest supported claim**: The Divisor Normalization Identity plus GWR leftmost minimum-divisor rule yields a unique structural certificate for every chamber reset in the 10^12–10^14 regime (see PROOF.md:1423 and research/02-gwr-dni/endpoint-01234.json).
**Explicitly not proved**: Closure of the modulus-link for all endpoint chains beyond 10^14; behavior under simultaneous multi-chamber resets.
**Failed and must not be revived**: Shadow selector (invalidated 2026-05-10, see research/02-gwr-dni/invalidation-log.md).
**Next session first action**: Run the boundary-drop admissibility probe on the new 10^14–10^15 slice using the pinned seed and the exact command in research/00-index/continuity/next-command.txt.
**Reproduce current state**: `python research/02-gwr-dni/validate_dni.py --slice 1e14-1e15 --seed 0xfeedface`
This skill is the immune system for long-running, high-precision research programs.
04 同仓库更多 Skills
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.
zoom-apps-sdk
Use when using Apps SDK.
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".
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.
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]/.
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".