@delorenj/pjangler 1.2.18 → 1.2.21

package/templates/commonproject/template/.mise/scripts/unlink-project-skills-from-clis.sh

@@ -1,45 +0,0 @@
-#!/bin/bash
-# Remove this project's skill symlinks from shared (global-scope) per-CLI dirs on
-# leave, so other projects / your global setup aren't polluted. Project-scoped
-# (local-scope) mirrors like ./.kimi-code/skills are left in place — they belong
-# to the repo and cost nothing to keep.
-#
-# Mirrors the SKILL_TARGETS table in link-project-skills-to-clis.sh.
-set -euo pipefail
-
-script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-project_root="$(cd "${script_dir}/../.." && pwd)"
-agents_skills="${project_root}/.agents/skills"
-
-SKILL_TARGETS=(
-  "${CODEX_HOME:-$HOME/.codex}/skills|global"
-  "${project_root}/.kimi-code/skills|local"
-)
-
-[ -d "$agents_skills" ] || exit 0
-
-_readlink_norm() {
-  local l; l="$(readlink "$1" 2>/dev/null || echo "")"
-  printf '%s' "${l%/}"
-}
-
-for entry in "${SKILL_TARGETS[@]}"; do
-  dir="${entry%%|*}"
-  scope="${entry##*|}"
-  [ "$scope" = "global" ] || continue
-  [ -d "$dir" ] || continue
-
-  unlinked=0
-  for skill_path in "$agents_skills"/*; do
-    [ -e "$skill_path" ] || continue
-    name="$(basename "$skill_path")"
-    [[ "$name" == ".system" ]] && continue
-    target="${dir}/${name}"
-    if [ -L "$target" ] && [ "$(_readlink_norm "$target")" = "$skill_path" ]; then
-      rm "$target"
-      unlinked=$((unlinked + 1))
-    fi
-  done
-  [ "$unlinked" -gt 0 ] && echo "Unlinked ${unlinked} project skill(s) from ${dir}"
-done
-exit 0

package/templates/hermes-agent/docs/bloodbank-gateway.md

@@ -1,57 +0,0 @@
-# Bloodbank Gateway
-
-Bloodbank command events should be a first-class Hermes gateway, not an inbox
-directory that a later heartbeat drains.
-
-## Goal
-
-Make this true:
-
-```
-Telegram message -> Hermes gateway -> agent turn queue -> agent acts
-Bloodbank command -> Hermes gateway -> agent turn queue -> agent acts
-Web message      -> Hermes gateway -> agent turn queue -> agent acts
-```
-
-The transport changes; the agent turn contract does not.
-
-## Current State
-
-`runtime-scaffold/bloodbank-consumer.py` subscribes to NATS subjects and writes
-JSON files under `bloodbank-inbox/`. That proves subscription, envelope parsing,
-and durable audit capture, but it is not command execution. It can accumulate
-files while the agent never acts.
-
-## Target Contract
-
-- The Bloodbank gateway subscribes to `bloodbank.cmd.v1.agent.<agent_id>.>`.
-- Each command event is converted into the same internal Hermes message/turn
-  shape used by Telegram and web gateways.
-- Hermes owns backpressure: if the agent is busy, the command waits in the
-  message queue until the agent can act.
-- The gateway emits lifecycle events:
-  - received
-  - accepted or rejected
-  - turn_started
-  - turn_completed or turn_failed
-- Replies use the incoming `correlationid` and `reply_to`/reply subject when
-  present.
-- Idempotency uses the CloudEvent `id`; duplicate events do not create duplicate
-  turns.
-
-## Required Work
-
-1. Add a Hermes transport adapter for Bloodbank/NATS.
-2. Define a command payload schema that maps cleanly to a user prompt plus
-   metadata (`actor`, `source`, `priority`, `correlationid`, `reply_to`).
-3. Replace the file-inbox consumer service with `hermes gateway bloodbank run`
-   or equivalent.
-4. Keep optional audit persistence, but make it observational only.
-5. Add smoke tests that publish a NATS command and assert that a Hermes turn is
-   created, processed, and acknowledged.
-
-## Non-Goals
-
-- No heartbeat inbox drain.
-- No polling directory as the command execution mechanism.
-- No per-agent one-off consumer logic.

package/templates/hermes-agent/docs/fleet-control-plane/n8n-service-hub.md

@@ -1,60 +0,0 @@
-# n8n Service Hub
-
-n8n should become the visible service hub for fleet operations while systemd
-remains the local survival layer.
-
-## Shape
-
-- systemd owns process supervision on each host.
-- n8n owns visibility, operator controls, scheduling, and multi-service flows.
-- pjangler owns deterministic generation and reconciliation of both sides.
-
-## Service Model
-
-Each installed Hermes agent service is represented as an n8n node/workflow item:
-
-- gateway service
-- Bloodbank gateway or transitional consumer
-- heartbeat timer/service
-- runtime checkpoint status
-- ticket-provider health check
-
-The node does not replace systemd. It calls stable commands such as:
-
-```
-systemctl --user status hermes-<agent>-heartbeat.timer
-systemctl --user restart hermes-<agent>-gateway.service
-pj fleet status --agent <agent_id> --json
-pj fleet reconcile --agent <agent_id> --apply
-```
-
-## Heartbeat v2
-
-Checkpoint era means the only recurring task was an hourly runtime
-commit/push. It preserved state but did not manage work.
-
-Heartbeat v2 is a frequent, cheap tick:
-
-- always performs a gated runtime checkpoint
-- optionally performs autonomous board reconciliation when
-  `.project.json automation.reconcile.enabled` is true
-- records status in runtime state files for fleet status readers
-- can be triggered by systemd locally or by n8n centrally
-
-## First Implementation Pass
-
-1. Extend `pj fleet status --json` to include service unit state, last heartbeat,
-   last checkpoint, and ticket-provider resolution.
-2. Generate an n8n workflow from `~/.hermes/agents-registry.yaml`.
-3. Add n8n actions for status, restart, reconcile dry-run, reconcile apply, and
-   log tail.
-4. Keep systemd timers installed and enabled as fallback.
-5. Add drift detection when n8n and systemd disagree about an agent.
-
-## Open Questions
-
-- Whether n8n should invoke local commands through SSH, an agent-side webhook,
-  or a small host-local control API.
-- How much service history should live in n8n versus Hermes runtime state.
-- Whether Bloodbank gateway events should also drive n8n workflows for visible
-  operator traces.