cianfhoghlaim/cianchosaint / .agents/skills/knowledge-sync-loop
knowledge-sync-loop
UnverifiedThe 6-layer pull-based sync architecture that keeps all 8 knowledge surfaces in the cianfhoghlaim repo in sync — openspec changes/specs, agent skills, MCP servers, code (CCC), docs (Cognee), Dagster, agent definitions, and the lakehouse. Use when the user asks 'how do I keep openspec in sync with the rest of the repo', 'how do I sync skills + MCP + CCC + Cognee', 'what's the knowledge graph of openspec changes', 'how do I grow the knowledge surface over time', or 'where is the sync health dashboard'. Per the 2026-08-15-knowledge-sync-loop-v1 change + the 2026-08-14-firecrawl-mcp-ccc-dual-search-v1 change (Layer 12 — Firecrawl CCC concept guides). Triggers: 'sync:paths', 'sync:ccc', 'sync:cognee', 'sync:skills', 'sync:mcp', 'sync:firecrawl', 'sync:all', 'stedding/sync-reports/', 'deployment control panel', 'notebook 24'.
Knowledge Sync Loop
The 5-layer pull-based sync architecture that keeps the 8 knowledge surfaces in the cianfhoghlaim repo in sync — and grows them over time via 3 feedback loops.
The 8 knowledge surfaces
The cianfhoghlaim repo has 8 knowledge surfaces that need to stay in sync:
- openspec changes (71 specs, 14 pending, 303 archived)
- openspec specs (71 capability specs)
- agent skills (57 SKILL.md files in
.agents/skills/) - MCP servers (14 servers wired in
opencode.json) - Dagster assets (~833 across the 5-layer defs/ tree)
- CCC chunks (257,957 / 8,845 files via CocoIndex Code)
- Cognee docs (1,743 .md files / 7 typed clusters)
- agent definitions (12-agent fleet + 8 NCCA subject specialists)
The 5-layer pull-based sync architecture
The new 2026-08-15-knowledge-sync-loop-v1 change establishes a
5-layer sync architecture with 6 mise run sync:* tasks + 1
orchestrator. Pull-based (per the user's choice) = no Dagster
sensors fire automatically, no GitHub Actions run on every push;
the developer (or the bring-up smoke test) invokes the tasks
explicitly.
| Layer | Task | What it does | Output |
|---|---|---|---|
| 1. Path | mise run sync:paths | Detect pre-v7 path drift in source files (6 patterns) | stedding/sync-reports/paths-{date}.md |
| 2. Index | mise run sync:ccc | Refresh CCC index + append the 20th concept guide | stedding/sync-reports/ccc-{date}.md |
| 3. Graph | mise run sync:cognee | Ingest openspec + skills into 3 new Cognee clusters | stedding/sync-reports/cognee-{date}.md |
| 4. Skills | mise run sync:skills | Run lint-skills.sh + validate_skill_references.py | stedding/sync-reports/skills-{date}.md |
| 5. MCP | mise run sync:mcp | List all 14 MCP servers from opencode.json | stedding/sync-reports/mcp-{date}.md |
| Orchestrator | mise run sync:all | Runs 1-5 in sequence | stedding/sync-reports/all-{date}.md |
Each task writes its report to stedding/sync-reports/ so the
deployment control panel (notebook 24) can consume it.
The 3 feedback loops (how the system grows over time)
Loop 1: Skill evolution
SKILL.md updated
→ sync:skills detects the change
→ lint:skills validates the frontmatter
→ validate_skill_references.py checks the references
→ Cognee cognifies the skill into the agent_skills cluster
→ CCC reindexes (the 20th concept guide "openspec-archive-search" can find it)
→ The deployment control panel surfaces the new skill
Growth trigger: When a skill's description: field changes, the
loop automatically re-cognifies it into Cognee. Over time, the
agent_skills cluster grows a knowledge graph of all 57+ skills +
their relationships.
Loop 2: openspec evolution
openspec change archived (e.g. via `openspec archive <id> --yes`)
→ sync:openspec-to-ccc updates the 20th concept guide
→ sync:openspec-to-cognee ingests the archived change
→ The Cognee graph adds the change as a new node
→ The deployment control panel shows the new "archived" status
Growth trigger: When an openspec change is archived, it automatically becomes part of the knowledge graph. Over time, the graph accumulates a complete history of every change + spec delta + implementation.
Loop 3: MCP evolution
MCP server config changes (in opencode.json)
→ sync:mcp detects the new server
→ Validates the server is reachable + responds to a healthcheck
→ Adds the server to the Cognee agent_skills cluster
→ The deployment control panel shows the new server
Growth trigger: When opencode.json is modified with a new MCP
server, the loop validates + cognifies it. Over time, the graph
keeps a complete inventory of all 14+ MCP servers + their health
state.
Quick routing
| If you want to... | Do this |
|---|---|
| Bring the repo's knowledge state up to date | mise run sync:all |
| Check just the path drift | mise run sync:paths |
| Check just the skills | mise run sync:skills |
| Read the latest sync report | cat stedding/sync-reports/all-$(date +%Y-%m-%d).md |
| See the canonical sync health | Open notebooks/24_deployment_control_panel.py in marimo |
| Add a new MCP server to the loop | Edit opencode.json + run mise run sync:mcp |
| Add a new openspec change to the loop | openspec archive <id> --yes + run mise run sync:all |
| Add a new skill to the loop | Edit .agents/skills/<slug>/SKILL.md + run mise run sync:skills |
Cross-references
openspec/changes/2026-08-15-knowledge-sync-loop-v1/— the change diropenspec/specs/knowledge-sync-loop/spec.md— the new specstedding/sync-reports/— the canonical per-layer reports dirscripts/sync/— the 6 standalone shell scriptsscripts/sync_openspec_to_ccc.py— the 20th concept guide appenderscripts/validate_skill_references.py— the skill path validatorscripts/cognee_ingest_openspec.py— the openspec Cognee ingesterscripts/cognee_ingest_skills.py— the skills Cognee ingesterorchestration/defs/sync_assets.py— the Dagster sync_health assetnotebooks/24_deployment_control_panel.py— the sync health dashboardopenspec/changes/2026-08-15-centralized-model-schema-registry-and-deployment-control-panel-v1/— Change B (the model-registry change that consumes the sync reports)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 318k | Updated | Claude CodeCodex | Unverified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 117k | Updated | Claude CodeCodex | Unverified | Install |