尚未收集儲存庫資訊。
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
製作者的介紹僅有上面的一句話。 更多詳情請閱讀儲存庫 README。 儲存庫 README ↗
正在載入完整 README。
檢視完整 README ↗A Delivery Record is a schema-typed, human-signed markdown file that documents one significant AI-assisted output — what was produced, what checks ran, and which named human reviewed it and what they actually verified. This repo ships the spec and the Agent Skills that keep evidence for, write, and verify records. The spec is browsable at https://kanopi.github.io/delivery-record/spec/v1.
AI can draft almost anything; accountability can't be delegated to it. The record is the artifact that proves a human stood behind the work: it refuses to exist without a named reviewer and two non-generic checkpoint notes.
spec/v1/ — the versioned spec: a JSON Schema for the record
front-matter, per-activity required-checks files, and a passing example for
every activity type. Fourteen activity types are covered, from code (PRs)
through frd, audit, discovery, design-handoff, strategy,
client-comm, design, qa, launch, deployment, devops,
project-setup, and ongoing-improvement.skills/delivery-record/ — the writing skill. Detects the activity
type, gathers facts (PR metadata, CI status, report paths), renders the
matching template, enforces the human checkpoint, writes the file, and
indexes it in the project's Teamwork "Delivery Records" notebook.skills/delivery-record-draft/ — writes an unsigned record so a
reviewer can read it. Leaves the reviewer and checkpoint notes blank, and
does not commit, push, or post to Teamwork. The draft fails verification on
purpose until it is finalized.skills/delivery-record-finalize/ — signs a draft after a human has
reviewed the work. Refuses without a named reviewer and two specific
checkpoint notes.skills/delivery-record-log/ — keeps an evidence log while the work
happens: a timestamped running record of the request, plan questions and
answers, decisions, findings, and checks, at
~/.claude/delivery-records/logs/<repo>/<branch>.md outside the working
tree. The writing skills read it when gathering facts. It never names a
reviewer.skills/delivery-record-verify/ — the read-only verification skill.scripts/delivery_record_verify.py — the validator both skills and CI
use: resolves the schema version from predicate_type, validates the
front-matter, and applies the threshold rule (a fail check needs a
## Waiver; an n/a check needs a one-line justification).
scripts/delivery-record-verify.sh is the CI wrapper.Claude Code (via the Kanopi marketplace):
/plugin marketplace add kanopi/claude-toolbox
/plugin install delivery-record@claude-toolbox
Local development:
git clone https://github.com/kanopi/delivery-record
claude --plugin-dir /path/to/delivery-record
Claude Desktop: download the plugin zip or individual .skill files from
the latest GitHub release and upload via Settings.
/delivery-record [--activity-type <type>] [--pr <n>] [--ticket <key>] [--reviewer <handle>]
/delivery-record-log [start|append|backfill] [<entry text>]
/delivery-record-draft [--activity-type <type>] [--pr <n>] [--ticket <key>]
/delivery-record-finalize [path] [--reviewer <handle>]
/delivery-record-verify [path] [--strict]
Use /delivery-record when the review has already happened.
Use /delivery-record-draft when it has not, then /delivery-record-finalize after the reviewer signs off.
Start /delivery-record-log early in a task that will need a record; the other skills read it and delete it once a signed record is written.
Or conversationally: "create a delivery record for this PR", "validate all the delivery records".
- name: Verify Delivery Records
run: scripts/delivery-record-verify.sh --strict
---
predicate_type: https://kanopi.github.io/delivery-record/spec/v1
activity_type: code
subject: {kind: pr, title: "Add breadcrumb layout", ref: "456", sha: "abc123"}
assisted_by: {models: ["claude-fable-5"]}
checks:
standards: {phpcs: pass}
tests: {unit: pass}
audits: {a11y: pass, performance: n/a}
review: {qa: pass}
sign_off:
produced_by: "@agent-session"
reviewed_by: "@jimbirch"
---
n/a: no performance-relevant changes in this PR.
## What the human verified
...
The full field reference, per-activity checks tables, and versioning policy
live in spec/v1/README.md and
spec/VERSIONING.md.
./scripts/validate-frontmatter.sh # frontmatter validation
bats tests/ # full test suite
node scripts/run-evals.js --min-rank1 75 # skill-routing evals
./scripts/check-codex-parity.sh # Codex artifact parity
Extracted from kanopi/cms-cultivator
— see MIGRATION.md for what changed, including the legacy
predicate_type alias that keeps pre-split records validating.
MIT © Kanopi Studios