change-publish

change-publish

What this skill does

Compile accepted ChDRs into the project's memory layer at {REPO_ROOT}/docs/adlc/memory/chdr/:

• Validate accepted ChDRs (status + provenance on Decision claims)
• Write each promoted ChDR to docs/adlc/memory/chdr/ChDR-{NNN}.md with OKF-style frontmatter
• Regenerate {REPO_ROOT}/docs/adlc/memory/chdr/chdr.md — the boot-facing index that team-boot injects into the session-start context (inside the class dir, same convention as adr/adr.md and pdr/pdr.md)
• Mark source drafts ### Status: **Published**

Unlike /team-levelup (which opens a PR against team-ai-directives), change-publish writes project-local memory — ChDRs describe this repo's evolution and fail the team-wide signal gate. No PR is created; the user commits via their normal flow.

This skill does not run until ChDRs have been accepted via /change-clarify.

When to use

• After /change-clarify: accepted ChDRs need promotion to memory
• After re-running /change-init: new accepted ChDRs to add to the memory index

When NOT to use

• No accepted ChDRs: run /change-clarify first
• Discovering ChDRs: use /change-init
• Reviewing ChDRs: use /change-clarify

Process

User Input

$ARGUMENTS

Examples:

• "ChDR-001 ChDR-003" — promote only specific ChDRs
• Empty input: promote all accepted ChDRs

Role & Context

You are acting as a Memory Publisher — moving accepted ChDRs from local drafts to durable project memory. Your role:

• Validate provenance one more time (the poisoning circuit breaker — a promoted unprovenanced decision poisons every future session via team-boot)
• Write promoted records with frontmatter that team-boot can index
• Regenerate the chdr.md index so the next session start sees the new decisions

Outline

1. Environment Setup (Phase 0): resolve paths, list accepted ChDRs
2. Prerequisites Check (Phase 1): ensure accepted ChDRs exist
3. Provenance Validation (Phase 2): re-verify every Decision claim has SHA/URL
4. Duplicate Check (Phase 3): skip ChDRs already in memory with same issue key
5. Memory Record Generation (Phase 4): write docs/adlc/memory/chdr/ChDR-{NNN}.md
6. Index Regeneration (Phase 5): rebuild docs/adlc/memory/chdr/chdr.md (boot-facing)
7. Draft Status Update (Phase 6): mark promoted drafts Published
8. Summary (Phase 7): report results

Execution Steps

Phase 0: Environment Setup

Run:

scripts/bash/setup-change-publish.sh

Parse JSON for REPO_ROOT, CHDR_DRAFTS_DIR, MEMORY_DIR, MEMORY_INDEX, ACCEPTED_CHDRS.

If the setup script is unavailable or fails, resolve manually:

1. REPO_ROOT — walk up to .adlc/, or git rev-parse --show-toplevel.
2. CHDR_DRAFTS_DIR — REPO_ROOT/.adlc/drafts/chdr
3. MEMORY_DIR — REPO_ROOT/docs/adlc/memory/chdr
4. MEMORY_INDEX — REPO_ROOT/docs/adlc/memory/chdr/chdr.md (inside the class dir, like docs/adlc/memory/adr/adr.md and pdr/pdr.md)
5. ACCEPTED_CHDRS — grep -l '^### Status: \*\*Accepted\*\*' CHDR_DRAFTS_DIR/ChDR-*.md

Phase 1: Prerequisites Check

If ACCEPTED_CHDRS is empty:

No accepted ChDRs found.
Run /change-clarify to accept ChDRs first.

Phase 2: Provenance Validation

For each accepted ChDR, re-verify: every non-trivial sentence in ### Decision references a SHA (\b[0-9a-f]{7,40}\b) or URL (https?://). Skip ChDRs that fail provenance — they cannot be promoted (poisoning risk):

## Provenance Validation

**Passing**: N | **Skipped**: M

### Skipped ChDRs

| ChDR | Reason |
|---|---|
| ChDR-XXX | Decision claim lacks SHA/URL provenance |

Skipped ChDRs remain Accepted in drafts for the user to fix.

Phase 3: Duplicate Check

For each passing ChDR, check if a memory record already exists with the same ### Issue Links key. If so, skip (or offer to merge) — do not create duplicates.

Phase 4: Memory Record Generation

For each accepted ChDR, write {MEMORY_DIR}/ChDR-{NNN}.md:

---
type: ChDR
title: {title from heading}
description: {descriptor from draft}
resource: ./docs/adlc/memory/chdr/ChDR-{NNN}.md
tags: [chdr]
generated:
  by: agent:change-publish
  at: {today}T00:00:00Z
id: ChDR-{NNN}
created: {date from draft}
verified:
  - by: agent:change-publish
    at: {today}T00:00:00Z
status: stable
stale_after: 365d
sources:
  - id: {sha}
    resource: git:{sha}
    title: {commit subject}
  - id: {issue-key}
    resource: {issue-url}
    title: {issue title}
---

# {Title}

{Content from draft — Context, Decision, Consequences, Evidence verbatim}

## Source

Promoted from: .adlc/drafts/chdr/ChDR-{NNN}.md

Phase 5: Index Regeneration (boot-facing)

This is the integration point with team-boot. Regenerate {MEMORY_INDEX} ({REPO_ROOT}/docs/adlc/memory/chdr/chdr.md) by listing all ChDR-*.md files in {MEMORY_DIR} and building a markdown table whose rows start with | ChDR- (the awk filter team-boot uses):

# Change Decision Records (Memory)

## ChDR Index

| ID | Title | Status | Date | Issues | Commits | Descriptor |
|----|-------|--------|------|--------|---------|------------|
| ChDR-001 | Why payments retries are capped at 3 | stable | 2026-08-16 | PROJ-123 | abc1234 | Consult before changing retry config |

**Stats**: N entries | Last Updated: YYYY-MM-DD

team-boot's boot.sh/boot.ps1 reads this file and emits a ## ChDR Index section + CHDR_COUNT into the session-start context, alongside the PDR/ADR indexes.

Phase 6: Draft Status Update

For each promoted ChDR, update the draft file's status to ### Status: **Published** and add:

### Promotion

- **Date**: [YYYY-MM-DD]
- **Memory path**: docs/adlc/memory/chdr/ChDR-{NNN}.md

Phase 7: Summary

## Change-Publish Summary

**ChDRs Promoted**: N
**ChDRs Skipped (provenance)**: M
**ChDRs Skipped (duplicate)**: K

### Artifacts

| Type | Count |
|---|---|
| Memory records (docs/adlc/memory/chdr/) | N |
| Boot index (docs/adlc/memory/chdr/chdr.md) | 1 (regenerated) |

### Next Steps

The next session start (`team-boot`) will inject the ChDR index into context.
Commit `docs/adlc/memory/chdr/` (records + `chdr.md`) via your normal flow.

Key Rules

Provenance Before Promotion

• A promoted unprovenanced Decision poisons every future session (team-boot injects it)
• Re-validate provenance at publish time, not just at clarify

Project-Local, Not Team-Wide

• ChDRs publish to docs/adlc/memory/, not team-ai-directives
• No PR created — user commits via normal flow
• ChDRs fail the team-levelup "team-wide applicability" signal gate by design

Index Format Must Match team-boot

• Rows MUST start with | ChDR- (the awk/regex filter in boot.sh/boot.ps1)
• File MUST be at docs/adlc/memory/chdr/chdr.md (inside the class dir, like adr/adr.md and pdr/pdr.md); dual-read fallbacks are docs/adlc/memory/chdr.md (older root-level placement) and .adlc/memory/chdr.md (legacy layout)

Workflow Guidance & Transitions

After /change-publish

/change-init → /change-clarify → /change-publish
    ↓
[team-boot] → injects docs/adlc/memory/chdr/chdr.md index at next session start
    ↓
[Agent consults ChDRs on demand when touching affected code]

Re-run /change-init periodically to mine new history; new accepted ChDRs are added to memory and the index is regenerated.

Next Steps

Commit the memory directory. The next session automatically sees the ChDR index via team-boot.

Verification

• All accepted ChDRs with provenance promoted to docs/adlc/memory/chdr/ChDR-*.md.
• docs/adlc/memory/chdr/chdr.md index regenerated with rows starting | ChDR-.
• Promoted drafts marked ### Status: **Published**.
• No unprovenanced ChDRs were promoted.
• No duplicate issue keys in memory.

Context

$ARGUMENTS