chore(state): clean up orphan gists + migrate routine state from gists to a dryvist/cc-routines-state repo

[JacobPEvans-personal]

Problem

The gists list at https://gist.github.com/JacobPEvans-personal is cluttered with 19 orphan state.json gists dated 2026-04-18 through 2026-05-06 — each 04:xx UTC, one per day, content {"last_polished": "...", "last_date": "..."}. These are from an old Daily Polish version that created a new gist every run instead of updating one. The current Daily Polish writes to daily-polish-state (single gist, properly named), so the 19 are dead.

Additionally, every cloud routine maintains its own *-state gist. Today there are ~4 named ones; once feat/six-new-routines lands there will be ~10 (apothecary-state, archivist-state, conductor-state, custodian-state, daily-polish-state, inspector-state, observer-state, quartermaster-state, sentinel-state, solver-state — plus the cross-routine routine-pr-budget). The gists UI becomes a wall of *-state rows that mean nothing visually.

Two-step fix

Step 1: Delete the 19 orphans (immediate, 1 command)

Requires a token tier with gist scope (e.g. gh-private or higher). From a properly-tiered shell:

gh gist list --limit 100 \
  | awk -F'\t' '$2 == "state.json" {print $1}' \
  | xargs -I{} gh gist delete {} --yes

Verify orphan count first:

gh gist list --limit 100 | awk -F'\t' '$2 == "state.json"' | wc -l
# expect: 19

The 5 named *-state gists (apothecary, daily-polish, issue-solver, weekly-scorecard, plus any others) are NOT affected — they have proper titles, not the bare state.json.

Step 2: Migrate state from gists to dryvist/cc-routines-state repo

Create a new private repo dryvist/cc-routines-state that holds one JSON file per routine. Routines read+write via createCommitOnBranch (the same signed-commit path the Solver already uses for target repos). Gists list stays clean forever.

One-time setup (user-only, requires gh-claude-dryvist and secrets-sync write)

# 1. Create the repo
gh repo create dryvist/cc-routines-state \
  --private \
  --description "State files for dryvist/claude-code-routines cloud routines. Each routine owns one JSON file. Routines read+write via createCommitOnBranch with JacobPEvans-claude App installation token."

# 2. Seed initial README (manual or via Contents API)
# 3. Add the new repo to the *github_app_repos anchor in
#    dryvist/secrets-sync/secrets-config.yml so the JacobPEvans-claude
#    App gets installation access.
# 4. Re-run secrets-sync distribution workflow.
# 5. Copy each current *-state gist content into:
#    dryvist/cc-routines-state:state/<routine>.json

Routine-prompt updates (one PR per routine, can be batched)

For each routine, replace the gist-read/write logic with Contents API equivalents against dryvist/cc-routines-state. Rough pattern:

# Read state
STATE=$(gh api repos/dryvist/cc-routines-state/contents/state/<routine>.json \
          --jq '.content' | base64 -d)

# Mutate state in /tmp/state.json...

# Write state via createCommitOnBranch (signed by JacobPEvans-claude[bot])
B64=$(base64 < /tmp/state.json | tr -d '\n')
jq -n --arg b64 "$B64" --arg sha "$BASE_SHA" '{
  query: "mutation($input: CreateCommitOnBranchInput!) { createCommitOnBranch(input: $input) { commit { oid } } }",
  variables: {
    input: {
      branch: { repositoryNameWithOwner: "dryvist/cc-routines-state", branchName: "main" },
      expectedHeadOid: $sha,
      message: { headline: "chore(state): update <routine> state [routine:<name>]" },
      fileChanges: { additions: [{path: "state/<routine>.json", contents: $b64}] }
    }
  }
}' | gh api graphql --input -

Routines in scope (10 total):

• On main today: daily-polish, sentinel, custodian, issue-solver (becoming solver), morning-briefing (becoming observer), weekly-scorecard (retiring)
• On feat/six-new-routines: apothecary, archivist, conductor, inspector, quartermaster (plus distributor which retires per chore(routines): retire The Distributor — replaced by org Required Workflows #28)

After migration: every routine deletes its legacy *-state gist on its first run after deployment, then writes to the new repo from there on.

Why a private repo, not a branch or another gist?

• No gist clutter. The whole point — gists list becomes just the user's actual personal gists.
• Browsable. GitHub repo UI is built for navigating files; gist UI is built for ephemera.
• Versioned. Every state mutation becomes a signed commit with a meaningful message. Auditable forever.
• Same mechanics routines already use. createCommitOnBranch is already in the Solver and (after feat/six-new-routines lands) everywhere else. No new auth path.
• Centralized backup. One repo to clone if you ever want to inspect or archive state.

A dedicated state-only branch in claude-code-routines was the runner-up option but bleeds branches into the source-of-truth repo. Separate repo is cleaner.

Out of scope

• The 5 legitimate personal gists (proxmox scripts, AI prompts, learning resources, Splunk references, everything-as-code) — those are user content, untouched.
• The routine-pr-budget cross-routine gist — same migration applies, gets its own file state/_pr-budget.json in the new repo.

Acceptance criteria

• 19 orphan state.json gists deleted; total gist count drops from 28 to 9 (5 personal + 4 routine state).
• dryvist/cc-routines-state private repo exists with README explaining purpose and JacobPEvans-claude App installed.
• Each of the 10 routine prompts updated to read+write state from the new repo (10 small PRs, OR one batched PR per branch).
• After all routines deploy: zero *-state gists exist; gh gist list shows only the 5 personal gists.

Suggested ordering

1. Delete orphans (1 command, immediate).
2. Create repo + README + secrets-sync wiring (one-time, ~15 min).
3. Migrate routine prompts in batches: first the 6 on main (covers Solver, Observer, Daily Polish, Sentinel, Custodian), then the 5 on feat/six-new-routines as part of that branch's merge.
4. Verify each routine's first post-deploy run deletes its legacy gist and writes to the new repo successfully.