agentleFS
OKHP3OKHP3/skillz/.agents/skills/okhp3-chatgpt-project-migration/SKILL.md

okhp3-chatgpt-project-migration

Skill3 starsChanged 2 months agoMITView on GitHub

OverKill Hill P³ ChatGPT project migration. Use when migrating, preserving, extracting, inventorying, normalizing, reconciling, or incorporating ChatGPT project threads, conversation exports, or complete chat histories into a repository. Also activate when a user needs a lossless thread archive, project-to-export crosswalk, provenance ledger, or public-safe distillation workflow. This is the authoritative capture and reconciliation process for ChatGPT project material in this repo -- use it even when the user asks only to "save", "review", or "summarize" a large set of threads.

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 8 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use a two-lane process:

    line 47
  2. Distinctive:

    Do not substitute artifact-lane summaries for source-lane records.

    line 54
  3. Distinctive:

    Use ChatGPT data export as the primary source when available.

    line 65
  4. Very specific:

    If the account cannot export data, proceed with Browser-based recovery but mark the batch ui_capture_only and list its reduced-fidelity …

    line 70
  5. Very specific:

    Create a disposable working copy under custom-gpts/ingestion/.

    line 83
  6. Distinctive:

    Do not call a migration complete until all are true:

    line 123
  7. Very specific:

    Every inventoried thread has a conversation_id and a disposition.

    line 125
  8. Very specific:

    Every captured source has a raw_sha256 and a normalized-record hash.

    line 126
  9. Very specific:

    Every unmatched thread, branch, message, or asset appears in exceptions.

    line 128
  10. Distinctive:

    Every committed artifact links to a source reference, a review decision, or an explicit statement that it is an original new artifact.

    line 130
  11. Distinctive:

    Use Notion only as an optional private Corpus Register for batch status, hashes, source locations, routing decisions, and review notes.

    line 151
  12. Distinctive:

    When processing already captured Custom GPT material, use the existing corpus ledger and evidence-register conventions.

    line 155
  13. Distinctive:

    When the user or verified target repository identifies an AskJamie or Glee-fully workflow, load only its bundled profile before applying …

    line 192

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

okhp3-chatgpt-project-migration

OverKill Hill P³ · overkillhill.com · github.com/OKHP3

Capture a ChatGPT project as evidence before transforming it. This skill keeps the source corpus intact, proves which threads and turns were captured, and only then routes reviewed material into durable repository artifacts.


Scope

In scopeOut of scope
Exporting and indexing accessible ChatGPT project threadsRecovering deleted or permission-restricted content
Preserving every exported message node and content partTreating an LLM summary as source preservation
Reconciling source material with repository artifactsPublishing raw chats, credentials, or third-party private data
Optional Notion status registerMaking Notion the authoritative raw archive

Operating principle

Use a two-lane process:

  1. Source lane: immutable export, project inventory, normalized transcripts, hashes, and exception records. Keep it local and out of Git.
  2. Artifact lane: reviewed, classified, public-safe Markdown and structured artifacts committed to their proper repository locations.

Do not substitute artifact-lane summaries for source-lane records. A source record may be distilled many times without changing the original evidence.


Required inputs and first decision

Record the project URL, capture date, account/workspace type, intended archive location, and publication sensitivity. Treat repository visibility as unknown unless the owner has resolved it.

Use ChatGPT data export as the primary source when available. It is faster and more complete than manually copying UI text. Use Browser to enumerate the target project and verify it against the export. Use Computer only for UI actions that Browser cannot complete, such as a download dialog or desktop-only control.

If the account cannot export data, proceed with Browser-based recovery but mark the batch ui_capture_only and list its reduced-fidelity limitations.


Capture procedure

1. Preserve the original

Store the downloaded export ZIP in an owner-controlled local archive outside the repository. Do not rename, edit, or unpack the only copy. Calculate and record raw_sha256 before working with it.

Create a disposable working copy under custom-gpts/ingestion/. That location is Git-ignored, but it is a staging area, not the permanent archive.

2. Inventory the project UI

Enumerate every accessible thread in the target project. For each thread record the stable conversation ID from its URL, displayed title, project URL, observed last activity, attachments or images, and UI-capture status. Do not use title matching as the primary key.

Write project_capture_manifest.json following references/source-manifest-schema.md. Assign each row one of: inventoried, captured, normalized, reconciled, routed, exception.

3. Normalize without flattening

Parse every conversation record into a normalized transcript and retain the raw JSON record. Preserve all nodes in the conversation mapping, including regenerated alternatives and non-current branches. Do not serialize only the current_node path.

For every message node retain: node ID, parent ID, child IDs, role, create and update time, author metadata, content parts in order, attachments, tool or citation metadata when present, and a source locator. Represent unavailable fields as null; do not invent values.

4. Reconcile and recover exceptions

Match project inventory rows to normalized export records by conversation ID. Record a one-to-one match, one-to-many ambiguity, or no match. Do not silently drop unmatched rows.

For every no_match or missing asset, use Browser-based capture as the recovery path. Preserve the URL, capture time, and reason for the exception. Keep the exception open until it has a source record or the owner accepts the loss.


Completion gates

Do not call a migration complete until all are true:

  • Every inventoried thread has a conversation_id and a disposition.
  • Every captured source has a raw_sha256 and a normalized-record hash.
  • The parser counted all conversation mapping nodes, not only visible turns.
  • Every unmatched thread, branch, message, or asset appears in exceptions.
  • Raw source material remains outside tracked repository paths.
  • Every committed artifact links to a source reference, a review decision, or an explicit statement that it is an original new artifact.

Report coverage as counts: threads inventoried/captured/reconciled, mapping nodes retained, assets retained, and open exceptions. Do not claim 100% unless the exception count is zero and the owner accepts the inventory boundary.


Integration and routing

Classify each candidate before writing to the repository:

ClassDestinationHandling
Raw evidenceLocal archive and ignored stagingPreserve verbatim; never commit by default
Historical referencecustom-gpts/proto/ or a clearly marked archiveRetain provenance and version context
Durable methodcustom-gpts/consolidated/, docs/, or a skillDistill, deduplicate, and cite source IDs
Operational artifactRelevant schema, example, skill, or child-repo materialAdd acceptance checks and ownership
Private or unsafe materialLocal quarantine recordRedact, omit, or obtain explicit owner approval

Use Notion only as an optional private Corpus Register for batch status, hashes, source locations, routing decisions, and review notes. Do not copy the entire source corpus into Notion merely to make it searchable.

When processing already captured Custom GPT material, use the existing corpus ledger and evidence-register conventions. Preserve disagreement and historical versions rather than making repetition appear authoritative.


Required handoff report

Return a concise report with:

Batch: <project name and capture date>
Source: export | ui_capture_only
Threads: <inventoried> inventoried / <captured> captured / <reconciled> reconciled
Nodes: <retained> retained
Assets: <retained> retained / <unavailable> unavailable
Exceptions: <count>
Artifact routing: <destinations and counts>
Owner decisions needed: <list or none>

References

  • references/source-manifest-schema.md -- capture manifest, normalized transcript, and reconciliation record contracts.
  • custom-gpts/ingestion/README.md -- local staging boundary for source files.
  • custom-gpts/consolidated/CORPUS-LEDGER.md -- existing evidence-ledger model.
  • custom-gpts/consolidated/EVIDENCE-REGISTER.md -- claim classification and source discipline.

Optional brand profiles

When the user or verified target repository identifies an AskJamie or Glee-fully workflow, load only its bundled profile before applying this shared procedure:

  • AskJamie: workflow profile.
  • Glee-fully: workflow profile.

For a neutral or unidentified target, use the shared procedure without a brand profile. An alias or previous package name alone does not establish the target brand. If the requested brand conflicts with the target evidence, resolve that conflict before applying a profile. Do not blend profiles or treat positioning as approved product canon. Profiles cannot weaken evidence, privacy, scope, or authorization requirements. Record the selected profile in the handoff.

Consolidation evidence

Version 1.1.0 bundles the former brand overlays as optional profiles as of 2026-09-19. Earlier evaluation designs and benchmark/results files retain their original versions and bytes as historical records; they do not validate this version or its profile selection. Profile regression cases are defined in the evaluation design and are not live runs. Structural validation is separate from behavioral or performance evidence.

About

Built by Jamie Hill · OverKill Hill P³ Published at github.com/OKHP3 Part of the OKHP3/skillz Agent Skill library. MIT License -- free to use, fork, and adapt. A nod to the source is appreciated.

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.