Skip to content

Jordanwei1/jiaojie-skill

jiaojie

Flagged

Create or receive human-first AI task handoffs across chats, models, devices, and languages. Use when the user asks to hand off, export context, resume from a handoff, switch sessions while preserving work, says "交接一下", "接收交接", or supplies handoff.md, handoff.zip, handoff-audit.zip, an LCH Bundle/T0 package, OCH Snapshot, or LTM Packet. Default to one readable handoff.md; upgrade to handoff.zip only when required files must travel, and to handoff-audit.zip only for formal audit, cross-organization delivery, or proof. Preserve current intent, stop point, next action, decisions, constraints, rejected and failed paths, answered questions, materials, omissions, and revalidation needs. Do not use for generic summaries, memory lookup, hidden reasoning, completed trivial chats, or after the user declines.

FileKindSize
SKILL.mdskill md8.1 KB
agents/openai.yamlconfig223 B
assets/hero-source.pngbinary1.3 MB
assets/hero.gifbinary1.2 MB
assets/profiles/materiality-v1.jsonconfig853 B
assets/protocol-version.jsonconfig2.6 KB
assets/registry/iana-language-subtag-registry-2026-08-08.txtmarkdown714.6 KB
assets/registry/iana-language-tag-extensions-registry-2014-04-02.txtmarkdown1.0 KB
assets/registry/profile-feature-registry-v0.1.jsonconfig3.3 KB
assets/registry/registry-lock.jsonconfig1.2 KB
assets/registry/schema-catalog-v0.1.jsonconfig4.1 KB
assets/schemas/action-graph.schema.jsonconfig5.6 KB
assets/schemas/common.schema.jsonconfig25.7 KB
assets/schemas/derived-digest-vector.schema.jsonconfig1.5 KB
assets/schemas/detached-envelope.schema.jsonconfig799 B
assets/schemas/event.schema.jsonconfig4.5 KB
assets/schemas/language-unicode-vector.schema.jsonconfig8.6 KB
assets/schemas/legacy-conversion-report.schema.jsonconfig4.4 KB
assets/schemas/manifest.schema.jsonconfig6.6 KB
assets/schemas/materiality-profile.schema.jsonconfig1.6 KB
assets/schemas/profile-feature-registry.schema.jsonconfig7.3 KB
assets/schemas/receipt.schema.jsonconfig15.6 KB
assets/schemas/registry-lock.schema.jsonconfig1.9 KB
assets/schemas/schema-catalog.schema.jsonconfig1.7 KB
assets/schemas/state.schema.jsonconfig11.9 KB
assets/schemas/t0-control.schema.jsonconfig2.7 KB
assets/schemas/vector-catalog.schema.jsonconfig5.3 KB
assets/schemas/verification-result.schema.jsonconfig24.2 KB
assets/templates/approval-statement.template.jsonconfig1.3 KB
assets/templates/approval-verification.template.jsonconfig1.6 KB
assets/templates/bundle-HANDOFF.template.mdmarkdown57 B
assets/templates/conversion-report.template.jsonconfig1.2 KB
assets/templates/handoff.template.mdmarkdown1.7 KB
assets/templates/manifest.template.jsonconfig8.6 KB
assets/templates/receipt.template.jsonconfig3.2 KB
assets/templates/t0-control.template.jsonconfig6.4 KB
assets/templates/t0-package.template.mdmarkdown337 B
assets/templates/warm-state.template.jsonconfig10.2 KB
assets/vectors/derived-digests-v1-001.jsonconfig10.8 KB
assets/vectors/index.jsonconfig2.1 KB
assets/vectors/jcs-sha256-basic-001.jsonconfig495 B
assets/vectors/language-unicode-v1-001.jsonconfig9.9 KB
assets/vectors/review-projection-v1-minimal-001.jsonconfig19.0 KB
examples/README.mdmarkdown1.4 KB
examples/business/business-external-state/case.jsonconfig559 B
examples/business/business-external-state/expected-next-action.mdmarkdown111 B
examples/business/business-external-state/expected-receipt.mdmarkdown342 B
examples/business/business-external-state/gold.jsonconfig498 B
examples/business/business-external-state/handoff.mdmarkdown1.8 KB
examples/business/business-external-state/source.mdmarkdown730 B
examples/coding/coding-cross-language/case.jsonconfig603 B
examples/coding/coding-cross-language/expected-next-action.mdmarkdown108 B
examples/coding/coding-cross-language/expected-receipt.mdmarkdown342 B
examples/coding/coding-cross-language/gold.jsonconfig495 B
examples/coding/coding-cross-language/handoff.mdmarkdown2.0 KB
examples/coding/coding-cross-language/source.mdmarkdown864 B
examples/general-chat/general-chat-mixed-constraints/case.jsonconfig577 B
examples/general-chat/general-chat-mixed-constraints/expected-next-action.mdmarkdown116 B
examples/general-chat/general-chat-mixed-constraints/expected-receipt.mdmarkdown347 B
examples/general-chat/general-chat-mixed-constraints/gold.jsonconfig508 B
examples/general-chat/general-chat-mixed-constraints/handoff.mdmarkdown1.8 KB
examples/general-chat/general-chat-mixed-constraints/source.mdmarkdown745 B
examples/index.jsonconfig4.9 KB
examples/learning/learning-intent-change/case.jsonconfig568 B
examples/learning/learning-intent-change/expected-next-action.mdmarkdown109 B
examples/learning/learning-intent-change/expected-receipt.mdmarkdown342 B
examples/learning/learning-intent-change/gold.jsonconfig496 B
examples/learning/learning-intent-change/handoff.mdmarkdown1.8 KB
examples/learning/learning-intent-change/source.mdmarkdown714 B
examples/product-design/product-design-parallel/case.jsonconfig572 B
examples/product-design/product-design-parallel/expected-next-action.mdmarkdown121 B
examples/product-design/product-design-parallel/expected-receipt.mdmarkdown342 B
examples/product-design/product-design-parallel/gold.jsonconfig508 B
examples/product-design/product-design-parallel/handoff.mdmarkdown1.8 KB
examples/product-design/product-design-parallel/source.mdmarkdown732 B
examples/research/research-partial/case.jsonconfig553 B
examples/research/research-partial/expected-next-action.mdmarkdown114 B
examples/research/research-partial/expected-receipt.mdmarkdown347 B
examples/research/research-partial/gold.jsonconfig506 B
examples/research/research-partial/handoff.mdmarkdown1.8 KB
examples/research/research-partial/source.mdmarkdown718 B
examples/writing/writing-authority/case.jsonconfig578 B
examples/writing/writing-authority/expected-next-action.mdmarkdown94 B
examples/writing/writing-authority/expected-receipt.mdmarkdown342 B
examples/writing/writing-authority/gold.jsonconfig481 B
examples/writing/writing-authority/handoff.mdmarkdown1.8 KB
examples/writing/writing-authority/source.mdmarkdown703 B
references/conformance.mdmarkdown42.5 KB
references/convert-legacy.mdmarkdown14.0 KB
references/domain-business.mdmarkdown3.3 KB
references/domain-coding.mdmarkdown3.4 KB
references/domain-general-chat.mdmarkdown3.2 KB
references/domain-learning.mdmarkdown3.2 KB
references/domain-product-design.mdmarkdown3.3 KB
references/domain-research.mdmarkdown3.3 KB
references/domain-writing.mdmarkdown3.2 KB
references/export-workflow.mdmarkdown14.9 KB
references/multilingual-workflow.mdmarkdown9.6 KB
references/profiles.mdmarkdown21.1 KB
references/protocol-core.mdmarkdown28.7 KB
references/receive-workflow.mdmarkdown23.2 KB
references/result-model.mdmarkdown22.3 KB
references/runtime-capabilities.mdmarkdown6.8 KB
references/security-boundary.mdmarkdown8.8 KB
references/simple-workflow.mdmarkdown7.0 KB
references/terminology.mdmarkdown13.3 KB
references/threat-model.mdmarkdown20.1 KB
references/verify-structure-workflow.mdmarkdown21.9 KB
references/wire-format.mdmarkdown44.0 KB
scripts/_vendor/lch/__init__.pyscript1.4 KB
scripts/_vendor/lch/canonicalize.pyscript7.6 KB
scripts/_vendor/lch/cli.pyscript10.2 KB
scripts/_vendor/lch/converters.pyscript46.0 KB
scripts/_vendor/lch/language.pyscript27.7 KB
scripts/_vendor/lch/package.pyscript56.8 KB
scripts/_vendor/lch/projection.pyscript22.7 KB
scripts/_vendor/lch/registry.pyscript7.9 KB
scripts/_vendor/lch/schema.pyscript19.5 KB
scripts/_vendor/lch/security.pyscript7.2 KB
scripts/_vendor/lch/simple.pyscript36.6 KB
scripts/_vendor/lch/util.pyscript14.7 KB
scripts/_vendor/lch/validate.pyscript128.4 KB
scripts/convert_handoff_md.pyscript207 B
scripts/convert_ltm.pyscript201 B
scripts/convert_och.pyscript205 B
scripts/handoff.pyscript190 B
scripts/pack_handoff.pyscript175 B
scripts/project_check.pyscript3.2 KB
scripts/scan_sensitive.pyscript183 B
scripts/validate_handoff.pyscript180 B
scripts/verify_handoff.pyscript188 B

references/convert-legacy.md

View on GitHub
# Convert Legacy Handoffs

Read this reference only for `CONVERT_LEGACY` from a supported legacy input into a native handoff package.

Do not use `RECEIVE` for legacy input. `RECEIVE` accepts native packages only.

Treat this file as operational guidance. Do not add wire fields beyond the frozen protocol and schemas.

## Contents

- [Supported first-version inputs](#supported-first-version-inputs)
- [Isolate the source](#isolate-the-source)
- [Detect and record the format](#detect-and-record-the-format)
- [Map deterministically](#map-deterministically)
- [Keep model-assisted mapping untrusted](#keep-model-assisted-mapping-untrusted)
- [Degrade without a deterministic parser](#degrade-without-a-deterministic-parser)
- [Apply format-specific mappings](#apply-format-specific-mappings)
- [Set conservative output claims](#set-conservative-output-claims)
- [Finish safely](#finish-safely)

## Supported first-version inputs

- Convert a plain `HANDOFF.md` only through its frozen headings and structures.
- Convert only the canonical OCH Snapshot v1 six-field Markdown format defined below.
- Convert only a Core Memory Packet (CMP) whose source-declared `ltm_version` is
  exactly `"0.1"` or `"0.2"` and whose complete object conforms to the corresponding
  first-party CMP schema plus the exact-version rule below.
- Refuse automatic mapping when format detection is unknown or conflicting.
- Accept a user `--format` override only as a declared outside-input choice, not as
  proof that the input conforms. It cannot bypass an exact OCH/LTM version or the
  exact HANDOFF-v1 class. The sole fallback is the frozen generic HANDOFF mode below.

Do not claim support for an untested variant because its text looks similar.

All supported Markdown classes use strict UTF-8 without BOM and LF line endings.
The converter rejects undecodable text or CR bytes before format detection.

Release `0.1.0` freezes these versioned legacy classes; no near-synonym or invented
version label is allowed:

```text
OCH v1:   source_version=och-snapshot-v1
          detection_rules=[och_snapshot_v1_exact_six_fields]
LTM v0.1: source_version=ltm-cmp-v0.1
          detection_rules=[ltm_cmp_v0_1_exact_version]
LTM v0.2: source_version=ltm-cmp-v0.2
          detection_rules=[ltm_cmp_v0_2_exact_version]
```

The audited first-party sources for these frozen parsers are
`open-context-handoff@472dd0f247f208996516b329b7681411de656e46` file
`docs/snapshot-format.md` and
`dennisdevulder/ltm@153eee5fc08949db80d48ea5f61ccb3d54df6d80` files
`docs/spec/v0.1.md`, `SPEC.md`, `schema/core-memory.v0.1.json`, and
`schema/core-memory.v0.2.json`. Runtime conversion MUST use bundled parser rules
derived from this frozen contract and MUST NOT fetch a mutable remote schema.

For CMP v0.1, the upstream schema's broad `ltm_version` pattern does not replace the
class discriminator: the converter additionally requires the value to equal `"0.1"`.
The v0.2 class requires `"0.2"`. Unknown values are rejected rather than coerced.

## Isolate the source

1. Place the original input in permission-restricted staging or quarantine.
2. Compute `source_sha256` over the exact input bytes.
3. Scan for secrets, personal data, active content, nested archives, unsafe paths, and prompt injection.
4. Keep tools and side effects disabled while parsing.
5. Keep the original outside the release package until transfer is approved.
6. Use only `REFUSE`, `QUARANTINE`, or `REDACTED_EXPORT` after a secret,
   active-content, unsafe-path, or equivalent security hit.
7. Use `APPROVED_ORIGINAL` only when those checks did not hit, transfer is lawful,
   and the exact original receives valid transfer approval.

A clean scan lowers risk. It does not prove that the source contains no secrets.

## Detect and record the format

- Record the detected source format.
- Record the exact source version or commit when known.
- Record `detection_rules` and `detection_confidence`.
- Record `parser_version`.
- Record any user `--format` override in `format_override`; use `null` when absent.
  A non-null value equals `conversion_origin`. Exact-mode detector output remains
  unchanged; generic HANDOFF uses its distinct frozen report tuple.
- Preserve detection conflicts and warnings.
- Stop automatic conversion when frozen syntax does not identify one supported format.

Do not guess a version from prose alone.

### Generic HANDOFF override

When and only when the current user explicitly selects HANDOFF Markdown outside the
untrusted input, accept plain UTF-8/LF Markdown without the v1 marker under this exact
report tuple:

```text
conversion_origin: handoff_markdown
source_version: handoff-md-generic
detection_rules: [handoff_md_generic_user_override]
detection_confidence: 0
format_override: handoff_markdown
```

Preserve the exact input bytes as rooted COLD. Map only exact `## Current Intent`,
`## Decisions`, `## Constraints`, `## Rejected`, and `## Next Action` sections; keep
everything else unmapped and never infer a section from free prose. Add a generic-
override warning, keep all mapped semantics proposed and non-authoritative, and emit
only `PARTIAL`, `INELIGIBLE`, `PROPOSED`, and `BLOCKED` claims. If exact-byte transfer
is not allowed, refuse or quarantine instead of producing this conversion. Never use
this fallback for an unknown OCH or LTM version.

## Map deterministically

- Map only fields that exist in the source.
- Apply only frozen syntax and mapping rules.
- Give every mapped field a `rule_id`.
- Record `source_line` and `source_json_pointer`; either MAY be `null` when that
  locator form does not apply.
- Record `extraction_method` and `evidence_refs`.
- Preserve unknown sections in `unmapped_sections`.
- Preserve `conflicts` and `warnings`.
- Create omissions for missing material information.
- Keep stable record IDs and `canonical_state_digest` identical for the same input and deterministic options.

Do not invent missing intent, authority, evidence, decisions, constraints, artifacts, or history.

## Keep model-assisted mapping untrusted

- Preserve unsupported natural-language sections verbatim.
- Use model assistance only to propose mappings.
- Record the model, prompt, version, and confidence.
- Mark proposed results as `PROPOSED/UNTRUSTED`.
- Keep them out of ACTIVE canonical state until the current user confirms them.
- Exclude model-assisted suggestions from deterministic repeatability claims.

Do not upgrade a plausible interpretation to `USER_CONFIRMED`.

## Degrade without a deterministic parser

- Preserve the original bytes in restricted staging or quarantine.
- Keep every natural-language mapping as `PROPOSED/UNTRUSTED`.
- Do not move a proposed mapping into ACTIVE canonical state.
- Do not create a structural verifier result or deterministic-repeatability claim.
- Produce only a `PARTIAL` and `INELIGIBLE` proposed draft when policy permits.
- Route formal conversion to a T3 runtime with the frozen parser when available.

Do not call model interpretation deterministic conversion.

## Apply format-specific mappings

For a plain `HANDOFF.md`:

- Extract only frozen headings and structures deterministically.
- Put unknown headings in `unmapped_sections`.
- Preserve all remaining natural language as source data or an untrusted proposal.

For an OCH Snapshot:

- Require exactly six `###` fields in this order: `WHAT WE ARE DOING`,
  `CURRENT STATE`, `COMPLETED`, `DECISIONS`, `CONSTRAINTS`, and `NEXT ACTION`. A single
  document title MAY precede the first field; it is exactly one nonempty `# ` H1
  line with only blank lines before the first `###` field and is not a seventh
  field. Other preamble, trailing prose, or headings are not part of the canonical
  Snapshot and make exact-class detection fail. Additional, missing, reordered,
  renamed, or duplicate fields are nonconforming. In particular, canonical OCH v1
  has no `REJECTED` field.
- Enforce the machine-checkable field shape: nonempty task/current-state/next-action
  bodies and bullet-list completed/decisions/constraints (exactly `- None.` when an
  empty list is declared). Preserve the complete original bytes. The first-party
  semantic rules that the task is exactly one sentence, the state is concise, and
  the next action is one concrete observable action require language/human judgment;
  the deterministic converter records them as unverified instead of claiming that
  syntax proves them.
- In the three list fields, the exact sentinel `- None.` represents an empty list and
  MUST NOT become a completion, decision, or constraint Record.
- Map `WHAT WE ARE DOING` to proposed intent, `CURRENT STATE` to source-declared
  state context, each `COMPLETED` item to a source-declared completion claim,
  `DECISIONS` to candidate decisions, `CONSTRAINTS` to proposed constraints, and
  `NEXT ACTION` to a blocked proposed next action. Do not infer rejected paths.
- The target axes are fixed: intent `lifecycle: PROPOSED`; current-state and
  completion claims `epistemic_basis: EXTERNAL_ASSERTED`,
  `verification: UNVERIFIED`, `temporal_validity: UNKNOWN`; decisions
  `lifecycle: CANDIDATE`;
  constraints `lifecycle: PROPOSED`, `compliance: UNKNOWN`; and next action
  `eligibility: BLOCKED`.
- Keep every mapped value below `USER_CONFIRMED` unless separate current-user
  evidence establishes it; the source document's own statement that it was reviewed
  is not verifier evidence.
- `detection_confidence: 1` means the six-field source class and version were
  identified deterministically. It is not a claim that every semantic OCH field rule
  or historical human-review assertion was independently verified.

For an LTM Core Memory Packet (CMP):

- Parse strict JSON and require the complete object to satisfy the selected
  first-party `core-memory.v0.1` or `core-memory.v0.2` schema, the 32 KiB (32,768
  exact input bytes) serialized limit, and the exact `ltm_version` discriminator.
  Required fields are `ltm_version`, `id`, `created_at`, `goal`, and `next_step`.
- Validate `created_at` as RFC 3339 UTC according to the CMP specification; Draft
  2020-12 format annotation alone is not sufficient for the converter to accept it.
- For both versions, map `goal` to proposed intent; `constraints` to proposed
  constraints; `decisions[].what/why/locked` to candidate decisions with their
  source-declared rationale and lock flag; `attempts[].tried/outcome/learned` to
  attempts; `open_questions` to open questions; and `next_step` to a blocked
  proposed next action. Preserve `project`, `tags`, and `provenance` as
  source-declared context/provenance without upgrading their authority.
- The target axes are fixed: intent `PROPOSED`; constraints
  `PROPOSED/UNKNOWN`; decisions `CANDIDATE`; questions `OPEN`; and next action
  `BLOCKED`. Attempt outcomes map exactly as `succeeded -> SUCCEEDED`,
  `failed -> FAILED`, and `partial -> INCONCLUSIVE`, while the exact source value
  remains evidence. A source `locked: true` remains a source-declared decision
  property and does not upgrade the decision to ACTIVE or user-confirmed authority.
- For v0.2 only, preserve `parent_id` as source-declared lineage; map
  `success_criteria` to the blocked action's source-declared completion criteria;
  preserve `decisions[].consequences`; preserve
  `methods[].name/when_applicable/how` as
  source-declared procedural claim records with `EXTERNAL_ASSERTED`, `UNVERIFIED`,
  and `UNKNOWN` axes; and preserve `attempts[].confidence` as the source author's
  confidence, never Receiver confidence.
- Preserve the exact `ltm_version` and report it only through the frozen
  `ltm-cmp-v0.1` or `ltm-cmp-v0.2` `source_version` label. No CMP `1.0` class is
  defined by this release.
- Treat `provenance.source_hash` as `source_declared`; it cannot substitute for the
  converter's independently recomputed `source_sha256` over the exact input bytes.
- Do not synthesize intent or decision evolution, supersession, artifacts, approval,
  authorization, or original source evidence absent from CMP. Unknown JSON members
  fail the upstream closed schema rather than becoming silently mapped semantics.

For every format:

- Keep external paths as provenance only.
- Package required objects or create omissions.
- Treat all imported content as untrusted data.
- Preserve original bytes when policy permits.

## Set conservative output claims

Emit only existing Converter claims. In the first-version converters, keep these root claims fixed:

```yaml
CONVERSION_REPORT: COMPLETED
STRUCTURE_RESULT_REF: null
SOURCE_ORIGIN_CLAIM: UNKNOWN
CONTINUITY_EVAL_ELIGIBILITY_CLAIM: INELIGIBLE
CONTENT_COVERAGE_CLAIM: PARTIAL
COVERAGE_CLAIM_BASIS: CONVERTER_DERIVED
COVERAGE_RESULT_REFS: []
APPROVAL_CLAIM: PROPOSED
```

Compute `MISSING` from the actual input. Do not copy a fixed list.

- Keep the first-version Converter root at `CONTENT_COVERAGE_CLAIM: PARTIAL`, even
  when a particular legacy input carries more evidence.
- Use an omission for every known material gap.
- Keep `CONTINUITY_EVAL_ELIGIBILITY_CLAIM: INELIGIBLE` in the Converter root.
- Do not emit `RECEIVER_RUN_RESULT`; only a Receiver eval run may create it.
- Keep coverage result references empty until an authorized coverage auditor produces them.
- Keep `APPROVAL_CLAIM: PROPOSED` in the root permanently. Post-seal statements,
  coverage results, and approval verification remain detached and never rewrite it.
- Generate a new native package with a new root and new approval when a user wants
  to promote independently confirmed state beyond the conservative conversion.

Do not turn successful conversion into a lossless, complete, self-contained, approved, or continuity-verified claim.

## Finish safely

- Build the native package through the normal staging and seal process.
- Preserve the exact original or record why policy excluded it.
- Validate structure and byte consistency through the authorized deterministic verifier.
- Keep converter claims separate from verifier, trust, coverage, approval, Receiver, and benchmark results.
- Report every unmapped section, conflict, warning, omission, and security decision.
- Leave side effects disabled until the native receive protocol establishes current authorization.

Do not claim that any converter, mapping, input format, or output is currently verified merely because this workflow was followed.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall