chain-studio/tools/today
flemming-it ce97300a12 feat(studio): daily Today-Hero proposal pipeline (v0.31.0)
Cross-store research (Apple, Play, Steam, Docker, VS Code,
Chrome Web Store, Flathub) consistently rewards editorial
curation over algorithmic recommendations — but manual
copywriting per release does not survive a solo-dev cadence.
This commit lands a daily-build pipeline so the Today-Hero
card stays fresh without operator hand-edits per release.

Pipeline shape (full design in docs/today-pipeline.md):

1. tools/today/collect.sh aggregates "what happened in the
   last 24 hours" across the F∆I monorepos: git log per repo,
   store-index seed.yaml diffs, architecture/system-gaps doc
   changes, Studio release tags, and (opt-in) audit-log
   highlights. Outputs plain text.

2. tools/today/propose.sh feeds the signal summary plus
   prompt.template.md to the operator's already-configured
   System-AI (Ollama default; OpenAI-compatible endpoints
   work via env-var override). Drafts N candidate stories as
   YAML files under ~/.fai/today/proposals/<date>/.

3. tools/today/accept.sh validates a chosen candidate against
   the today/v1 schema and the no-marketing-speak banned-word
   list, then atomic-renames it into ~/.fai/today/active.yaml.

4. Studio reads active.yaml at store-page init via the new
   TodayStoryLoader (lib/data/today_story_loader.dart). On any
   failure (file missing, schema mismatch, banned-words hit,
   parse error) it falls back to the compiled-in
   _kFallbackTodayStory so KRITIS deployments and fresh
   installs always render something sensible.

Trust + audit:
- All proposed and accepted stories live as plain YAML on disk.
- The pipeline calls only the operator's already-configured
  System-AI; it never reaches a CMS, never phones home, works
  air-gapped if the System-AI does.
- The bash accept gate AND the Dart loader both enforce the
  banned-word list — a hand-edited active.yaml that bypassed
  the shell still won't reach the UI.
- Removing the cron entry disables the pipeline; Studio falls
  back to the const story and continues to work.

Cron / launchd / systemd recipes documented in
tools/today/README.md.

Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
2026-05-08 12:59:33 +02:00
..
accept.sh feat(studio): daily Today-Hero proposal pipeline (v0.31.0) 2026-05-08 12:59:33 +02:00
collect.sh feat(studio): daily Today-Hero proposal pipeline (v0.31.0) 2026-05-08 12:59:33 +02:00
prompt.template.md feat(studio): daily Today-Hero proposal pipeline (v0.31.0) 2026-05-08 12:59:33 +02:00
propose.sh feat(studio): daily Today-Hero proposal pipeline (v0.31.0) 2026-05-08 12:59:33 +02:00
README.md feat(studio): daily Today-Hero proposal pipeline (v0.31.0) 2026-05-08 12:59:33 +02:00

Today-Hero proposal tooling

Daily-build pipeline for the editorial Today-Hero card in Studio's store. See ../../docs/today-pipeline.md for the full design rationale.

Quick start

# 1. Generate today's proposals (calls the local Ollama by default).
./propose.sh

# 2. Skim them — they sit under ~/.fai/today/proposals/<ISO-DATE>/
ls ~/.fai/today/proposals/

# 3. Accept one. The chosen file becomes ~/.fai/today/active.yaml,
#    which Studio reads at startup.
./accept.sh ~/.fai/today/proposals/2026-05-08/candidate-2.yaml

# 4. Restart Studio (or let the next launch pick it up).

Schedule it

macOS — launchd

# Drop a plist that runs propose.sh every morning at 07:30 local time.
# `launchctl load ~/Library/LaunchAgents/ai.flemming.fai.today.plist` to
# install. The plist must point at the absolute path of propose.sh.

A starter plist lives at launchd/ai.flemming.fai.today.plist.

Linux — systemd timer

# Copy systemd/fai-today.{service,timer} into ~/.config/systemd/user/,
# then `systemctl --user enable --now fai-today.timer`.

Anywhere — cron

30 7 * * * /Users/flemming/.../tools/today/propose.sh > /tmp/fai-today.log 2>&1

Provider override

propose.sh defaults to Ollama at http://127.0.0.1:11434 with the model gemma3:4b. Override via env vars:

Variable Default Example
FAI_TODAY_API http://127.0.0.1:11434/api/generate https://api.openai.com/v1/chat/completions
FAI_TODAY_MODEL gemma3:4b gpt-4o-mini
FAI_TODAY_KEY (unset) $OPENAI_API_KEY
FAI_TODAY_N 3 1 (single shot)

OpenAI-compatible endpoints will work as long as they speak chat/completions or generate; the script auto-detects from the URL path.

Trust model

  • Reads only git log, git diff, and the bundled seed.yaml of installed F∆I monorepos. Never your audit log without explicit op-in.
  • Calls only the System-AI you've already configured for Studio. The same privacy mode you set there applies here.
  • Writes only into ~/.fai/today/. Nothing in ~/.fai/data/ or ~/.fai/config.yaml is touched.
  • Studio loads active.yaml at startup; if the file is missing or fails schema validation, the compiled-in fallback story renders. KRITIS fresh installs see the fallback until and unless an operator accepts a proposal.