draft-publish-team

name	draft-publish-team
description	Publishes the curator's local context to the shared GitHub repo. Reads change log entries since last publish, builds CHANGES.jsonl, pushes to the team repo via the separate-clone pattern. Run whenever you want teammates to see updated context.

/draft:publish-team — Publish Context to Shared Repo

Push your local product context to the shared team repo so teammates can load it.

Flags: --no-confirm — skip the preview step (used by /draft:setup-collab auto-seed and automation).

---

Step 1: Read config and verify auth

Read $DRAFT_WORKSPACE/config/collaboration.json using json.loads(). Read $DRAFT_WORKSPACE/config/local.json using json.loads().

If config/collaboration.json is missing or mode is not github:

Error: Collaboration not configured. Run /draft:setup-collab first.

Hard stop.

Verify gh auth:

gh auth status

If fails:

GitHub auth expired. Run `gh auth login --web` then re-run /publish-team.

Hard stop.

---

Step 1.5: Apply pending synthesis proposals from proposals/

Check for .md files in $DRAFT_WORKSPACE/proposals/.

If proposals/ is empty or doesn't exist: skip to Step 2.

If proposals/ has files:

1. List the files:

   ls "$DRAFT_WORKSPACE/proposals/"*.md 2>/dev/null
   

2. For each .md file, read and display it to the user:

   Pending synthesis proposals:
   ─────────────────────────────────────────────────────
   File: 20260517T024408Z-853ea41f.md
   Session: 853ea41f | Source: session | Synthesized by: claude-code
   
   [markdown body content — the human-readable preview section]
   ─────────────────────────────────────────────────────
   

3. Ask the user: "Apply all synthesis proposals to context before publishing? [Y/n/skip-all]"

   • Y or Enter → apply all
   • n → show each proposal individually and ask per-file (y/skip/delete)
   • skip-all → skip proposals/ entirely for this publish (files stay for next time)
   • delete-all → archive all proposals to proposals/rejected/ without applying

4. Applying a proposal: Parse the YAML frontmatter context_updates and apply each update:

   • action: append → append content to the end of $DRAFT_WORKSPACE/<file>
   • action: tension → append content to $DRAFT_WORKSPACE/context/tensions.md regardless of the file: field value (the file: field describes which dimension the contradiction relates to, not where to write)
   • action: overwrite → replace the file contents with content (curator-triggered compaction only — synthesizers should never produce this action; if present in a daemon-synthesized proposal, it indicates a prompt guardrail failure and commit-to-team-context.sh will reject it)
   • Create the target file if it doesn't exist
   • Log each applied update
   • Per-file delete → archive to proposals/rejected/ (do not apply):

     mv "$DRAFT_WORKSPACE/proposals/<file>.md" "$DRAFT_WORKSPACE/proposals/rejected/<file>.md"
     

5. After applying: note the session_id, synthesized_by, and input_source from each file's frontmatter — you will need these for Step 4 CHANGES.jsonl entries.

6. After successful push (Step 4): archive applied proposals/ files to accepted/:

   mkdir -p "$DRAFT_WORKSPACE/proposals/accepted" "$DRAFT_WORKSPACE/proposals/rejected"
   mv "$DRAFT_WORKSPACE/proposals/<applied_file>.md" "$DRAFT_WORKSPACE/proposals/accepted/<applied_file>.md"
   

   Only archive files that were successfully applied AND pushed. For delete-all: archive all pending files to rejected/ instead:

   mv "$DRAFT_WORKSPACE/proposals/"*.md "$DRAFT_WORKSPACE/proposals/rejected/"
   

---

Step 2: Collect changes

Read all files in $DRAFT_WORKSPACE/context/*/log/ with modification timestamps newer than config/local.json:last_published.

• If last_published is null (from local.json): this is the first manual publish — read ALL log entries.
• If no log/ directories exist or they are empty: no log-based entries.

Case: last_published is null AND no log entries exist

First publish with no history. Generate synthetic CHANGES entries — one per context dimension where index.md description is not "No information recorded yet":

For each qualifying dimension, compute the ID deterministically via bash:

ts=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
gh_username=$(gh api user --jq .login)
dimension="[company|product|team|priorities|decisions|tensions]"
file="context/${dimension}/index.md"
id=$(echo -n "${ts}${gh_username}${dimension}${file}" | sha256sum | cut -c1-8)

Generate CHANGES entry:

{"id":"[computed]","ts":"[current ISO 8601]","author":"[gh_username]","dimension":"[dim]","summary":"Initial context snapshot","file":"context/[dim]/index.md","log_entry":null}

Case: no entries found since last_published

Print: "Nothing new to publish since [last_published]." Exit 0.

IMPORTANT: Never compute IDs mentally. Always run the sha256 bash command above.

---

Step 3: Preview (skip if --no-confirm)

Show:

Publishing to: [team_repo_url] / [team_repo_subdir]
Changes:
  [N] entries across [dimensions list]
  [list of summaries, truncated at 80 chars each]
Teammates: [list from collaboration.json]

Confirm? [Y/n]

If n or N: "Aborted. Nothing published." Exit 0.

---

Step 4: Execute — separate-clone pattern

The Draft workspace (~/.draft/workspace) is NEVER initialized as a git repo. All git operations run in a short-lived temp directory.

TMPDIR=$(mktemp -d /tmp/draft-publish-XXXX)
git clone [team_repo_url] "$TMPDIR"

Set the subdir prefix. If team_repo_subdir is root, use no prefix (write to repo root):

# team_repo_subdir = root → SUBDIR_PATH="$TMPDIR"
# team_repo_subdir = .draft → SUBDIR_PATH="$TMPDIR/.draft"
# team_repo_subdir = [custom] → SUBDIR_PATH="$TMPDIR/[custom]"

Create target directories:

mkdir -p "$SUBDIR_PATH/context"
mkdir -p "$SUBDIR_PATH/config"

Copy context from workspace:

cp -r "$DRAFT_WORKSPACE/context/" "$SUBDIR_PATH/context/"

personal/ is structurally excluded — it lives outside context/ and is never under the copy source path.

Copy collaboration config:

cp "$DRAFT_WORKSPACE/config/collaboration.json" "$SUBDIR_PATH/config/collaboration.json"

Build CHANGES.jsonl:

1. Read existing $SUBDIR_PATH/CHANGES.jsonl if it exists (preserves history).

2. For each new CHANGES entry, compute ID via bash:

   id=$(echo -n "${ts}${author}${dimension}${file}" | sha256sum | cut -c1-8)
   

3. Append new entries (one JSON object per line). Two entry types:

   Manual publish (from log/ entries):

   {"id":"[computed]","ts":"[ISO8601]","type":"manual-publish","author":"[gh_username]","dimension":"[dim]","summary":"...","file":"context/[dim]/index.md","log_entry":"..."}
   

   Daemon synthesis (from proposals/ proposals applied in Step 1.5): For each applied synthesis file, compute ID the same way (sha256 of ts+author+source+session_id):

   id=$(echo -n "${ts}${gh_username}${input_source}${session_id}" | sha256sum | cut -c1-8)
   

   {"id":"[computed]","ts":"[ISO8601]","type":"daemon-synthesis","source":"[input_source]","session_id":"[session_id]","synthesized_by":"[synthesized_by]","approved_by":"[gh_username]","files":["context/product/index.md",...]}
   

4. Write back to $SUBDIR_PATH/CHANGES.jsonl.

Generate semantic commit message. If synthesis proposals were applied, note them:

"[N] context updates: [dimension list]"                   # manual only
"[N] context updates + [M] synthesis proposals applied"   # with synthesis

Stage and commit:

# Adjust git add paths based on team_repo_subdir (root vs subdir)
git -C "$TMPDIR" add "[subdir_relative]/context/" "[subdir_relative]/config/" "[subdir_relative]/CHANGES.jsonl"
git -C "$TMPDIR" commit -m "[semantic message]"
git -C "$TMPDIR" push

If push fails:

rm -rf "$TMPDIR"

Print the error. DO NOT update config/local.json.

Publish failed. Check your network and repo access, then try again. Nothing was changed locally.

Exit 1.

rm -rf "$TMPDIR"

---

Step 5: Update local state (success only)

Update $DRAFT_WORKSPACE/config/local.json:

• Read the file with json.loads(), set last_published to the current ISO 8601 timestamp, write back with json.dumps().

Print:

Published [N] changes to [team_repo_url]