Skip to content

PRO-259: adapt delivery loop for Hermes and Telegram - #6

Open
bluestonetrust wants to merge 1 commit into
finna:mainfrom
bluestonetrust:agent/pro-259-hermes-loop
Open

PRO-259: adapt delivery loop for Hermes and Telegram#6
bluestonetrust wants to merge 1 commit into
finna:mainfrom
bluestonetrust:agent/pro-259-hermes-loop

Conversation

@bluestonetrust

Copy link
Copy Markdown

What changed

Reworked the starter into a Hermes-native delivery loop:

  • replaced the three Finn/Claude/Linear skills with hermes-spec, hermes-build, hermes-review, and hermes-loop;
  • made Multica the only queue and durable contract/status source;
  • made Telegram the human alert and control surface;
  • added fresh commit-bound review, durable repair-round markers, a hard two-round cap, stuck escalation, and human-only merge;
  • rewrote installation and operating documentation for Hermes Agent;
  • expanded validation to enforce 17 critical workflow and safety contracts and reject stale workflow terminology.

Closes PRO-259

Scope ledger

  • Telegram replaces the previous chat control surface for blocked, stuck, and merge-ready alerts.
  • Multica replaces the previous issue queue and approval flow.
  • GitHub remains the source of truth for branches, pull requests, checks, review evidence, and merge history.
  • Repair convergence survives session restarts through durable PR comments named Hermes Repair round N/2.
  • Independent review is required; no independent context means human review.
  • Agents never merge, auto-merge, deploy, publish, spend, or mutate production.
  • Other behavior changes: None.

Verification

  • node scripts/validate.mjsValidated 4 Hermes skills, README links, and 17 safety contracts.
  • node --check scripts/validate.mjs → passed.
  • git diff --check origin/main...HEAD → passed.
  • CLI contracts checked against live hermes send --help, hermes skills install --help, multica issue create --help, multica issue pull-requests --help, and gh pr checks --help.
  • Independent review found three must-fix gaps. All were addressed: durable round counting, a complete project-bound Multica create command, and stronger validation assertions.

Manual verification

  1. Run node scripts/validate.mjs and confirm all four skills plus 17 safety contracts validate.
  2. Inspect each skill frontmatter and confirm its name matches its directory.
  3. Confirm the README installation commands point to all four raw skill files.
  4. Confirm hermes-loop stops before a third repair and sends Telegram alerts only after durable state is written.

Risk: Medium. This replaces the documented workflow and shipped skills, but does not alter application runtime code or merge automatically.

Co-authored-by: multica-agent <github@multica.ai>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant