← Back to .knowledge

// Blog / open source discovery

Open-source agent context layer for coding teams

For coding teams, an agent context layer should be inspectable, versionable, and honest about what the agent should not trust.

// direct answer

Short answer

An open-source agent context layer should keep routing, trust, freshness, evidence, repair work, and local tools in the repository so the team can inspect the agent's operating context alongside the code.

Evaluate whether an open core context layer is a better foundation than private prompt memory or hosted-only context.

// real failure mode

The failure mode

Agent context often becomes a private mix of chat history, local scratch files, hosted memory, and one-off prompts.

That makes it hard for a team to review what the agent is relying on, especially when multiple tools touch the same repo.

For coding teams, the context layer should be inspectable, versionable, and able to mark low-trust material instead of hiding it.

// repo-local model

The repo-local fix

.knowledge keeps the core context layer in the repository: routing, module cards, evidence, wiki, repair queue, metrics, and workflows.

The open core runs locally and does not require a hosted service for setup, release checks, search, or the static Inspector.

Commercial Pro2Pilot work can sit above that layer: team Inspector views, multi-repo dashboards, policy overlays, and advanced PR impact workflows.

// concrete example

Where GitHub, .knowledge, and Inspector fit

GitHub keeps the code, issues, PRs, and review history. .knowledge keeps a repo-local operating layer that agents can read before planning.

The local Inspector makes the open core visible as a static dashboard. Pro2Pilot Inspector Pro is the paid workflow layer for teams that want visual repair queues, policies, and PR impact review across real work.

// repo proof

Repo proof to inspect

Agent integrations

.knowledge/agent-integrations/

The same local context can be exposed to Codex, Claude Code, OpenCode, and command workflows.

Use it for
  • Multi-agent consistency
  • Repeatable onboarding
Do not use it for
  • Locking the team into one chat product
Local tools

.knowledge/tools/

Import, release, metrics, search, graph, and inspector commands are inspectable code.

Use it for
  • Auditing how reports are generated
  • Running checks in local or CI workflows
Do not use it for
  • Treating generated output as permanent truth
CI templates

.knowledge/github-action-templates/

The open core can be copied into repository automation without requiring a hosted control plane.

Use it for
  • PR health checks
  • Release evidence
Do not use it for
  • Assuming CI replaces human review
Static Inspector

.knowledge/inspector/

A local read-only dashboard makes trust, graph, and repair state visible.

Use it for
  • Visual sanity checks
  • Team discussion around stale modules
Do not use it for
  • Editing code from the dashboard
  • Bypassing source-of-truth rules

// command transcript

Commands and expected checks

node .knowledge/tools/flow.js release --no-color
expected
The open core regenerates routing, trust, metrics, graph, search, and Inspector outputs.
inspect next
.knowledge/maintenance/quality_report.json
node .knowledge/tools/serve-inspector.js
expected
The local static Inspector can be viewed from the generated artifacts.
inspect next
.knowledge/inspector/index.html
node .knowledge/tools/generate-pr-summary.js
expected
A PR-facing summary of trust, health, and repair state is generated.
inspect next
.knowledge/maintenance/pr_summary.md
Repo tree repo-local proof

What ships in the open core

01 agent-integrations/ -> Codex, Claude Code, OpenCode
02 docs/cookbook/ -> operational recipes
03 github-action-templates/ -> CI workflows
04 tools/ -> scan, import, release, metrics
05 inspector/ -> local read-only dashboard

// field guide

Open core vs paid workflow layer

The boundary should be explicit so the open source layer remains useful by itself.
Open corePaid Pro2Pilot layer
.knowledge routing, trust, freshness, evidence, and local tools.Team dashboards, policy overlays, and advanced review workflows.
Static local Inspector baseline.Interactive Inspector Pro with repair boards and PR impact views.
GitHub Action templates and local command workflows.Commercial support, team operations, and paid build delivery.

// guardrails

What the agent should not trust blindly

  • Open source matters here because teams can inspect the instructions, commands, generated reports, and trust rules.
  • Local-first does not mean isolated from CI. GitHub Action templates can run health, PR summary, and evaluation workflows after they are copied into a repository.
  • The static Inspector is a read-only view over .knowledge artifacts. Current code and tests remain the source of truth.

// common mistakes

Common mistakes

  • Confusing open-source context with hosted memory that only one vendor can inspect.
  • Hiding the free vs paid boundary until the team is already dependent on the workflow.
  • Using a visual dashboard as if it were the source of truth instead of a view over repository files.

// quick FAQ

FAQ

Is the open core useful without Pro2Pilot Inspector Pro?

Yes. The open core contains repo-local routing, trust, freshness, tools, integrations, metrics, and a static local Inspector baseline.

Why keep agent context in GitHub?

Because coding teams already review code, tests, docs, and automation there. Keeping agent context in the repo makes it inspectable and versionable.

// page-specific next step

Use this page when evaluating open agent infrastructure

Start with the GitHub repository and the local .knowledge layer. Add Pro2Pilot Inspector when the team needs visual operations on top of the open core.