context-archiving
Plan and execute hot-to-cold processkit context archiving while keeping archived entity metadata queryable through the shared index.
- License 许可
- MIT
- 最近更新
- 2026-11-11
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- context-archiving
- metadata
- { "processkit": { "id": "SKILL-context-archiving", "uses": [ { "skill": "index-management", "purpose": "Query entity metadata before selecting archive candidates." }, { "skill": "event-log", "purpose": "Record archive execution events when the write path exists." } ], "layer": 2, "created": 1777507200000, "version": "0.2.0", "category": "processkit", "provides": { "assets": [], "mcp_tools": [ "describe_archive_policy", "plan_archive", "create_archive", "extract_archive_payload" ], "primitives": [] }, "apiVersion": "processkit.projectious.work/v2" } }
- description
- Plan and execute hot-to-cold processkit context archiving while keeping archived entity metadata queryable through the shared index.
02 资源文件 2
- 脚本mcp/server.py11.6 KB
- 脚本scripts/test_context_archiving.py5.9 KB
03 SKILL.md 正文
Context Archiving
Intro
context-archiving manages the cold tier of processkit context. It identifies
stable archive candidates, packages their Markdown payloads into cold storage,
removes the hot files, and leaves metadata discoverable through
index-management.
Overview
Use this skill when context size is becoming expensive, old completed work is rarely read, or log/event payloads need to move from the hot tree into a cold archive while metadata remains queryable.
The cold-tier target shape is:
- Select stable entities by kind, state, and age.
- Package payload files into a content-addressed
tar.gzarchive. - Keep index rows queryable with
storage_locationpointing at the archive. - Extract full payloads on demand for historical queries.
Version 0.2.0 implements this shape with a dry-run-first archive writer and
explicit payload extraction tool.
MCP Tools
| Tool | Purpose |
|---|---|
describe_archive_policy() |
Return the default hot/warm/cold policy. |
plan_archive(kind, state, older_than_days) |
Return candidate entities from the live index. |
create_archive(kind, state, older_than_days, limit, dry_run) |
Write a tar.gz archive plus manifest, remove hot files, reindex, and log the archive event. Defaults to dry-run. |
extract_archive_payload(entity_id) |
Resolve an archived entity and return its full Markdown payload from cold storage. |
Default Policy
- WorkItems: archive terminal states after 30 days.
- DecisionRecords: keep accepted/superseded decisions hot for 90 days.
- LogEntries: shard by month; archive payload shards after 90 days.
- Migrations: archive applied or rejected history after 30 days.
- Never archive active, proposed, blocked, or in-progress entities.
- Never archive files under
context/templates/.
Gotchas
- Do not move entity files by hand. The index must keep metadata queryable
after archiving, so any write path has to update
storage_locationtransactionally. - Do not archive active state. Backlog, in-progress, blocked, proposed, and active entities belong in the hot tier.
- Treat logs as append-only. A cold log shard can be packaged, but its event identity and timestamp metadata must remain queryable forever.
- Do not trust filesystem age alone. Use entity metadata timestamps and state from the index; git checkout times are not archival age.
- Keep extraction explicit. A historical query may need full payload text, but routine status and routing queries should stay metadata-only.
Full reference
create_archive() writes two files under context/archives/YYYY/MM/: a
compressed tar payload and a JSON manifest. The manifest keeps entity metadata,
the original hot-tree path, the tar member path, a SHA-256 digest of each
payload, and the storage_location used by index-management.
After writing the manifest, the tool removes the archived hot files, rebuilds
the index, and records a context_archive.created event. dry_run=True is the
default and returns the exact candidate set without moving files.
Archived entities remain visible through normal index lookups. Use
extract_archive_payload(entity_id) only when the full historical Markdown is
needed.
04 同仓库更多 Skills
standard-ui-design
暂无简介
standard-agent-use
Company standard for working with AI agents. Use when deciding what to delegate to agents, capturing open work, keeping agent memory or knowledge notes, improving an agent workflow, or introducing agent work in a team or for a client.
standard-agent-building
暂无简介
skill-management
Manage Skill package manifests as v1 ontology entities.
schema-management
Manage generated processkit schema contracts through deterministic MCP rebuild and inspection workflows.
proposition-management
Record claims, risks, beliefs, world facts, WSJF estimates, and assumptions through the Proposition interface.