The Product Surface

What shapes the surface

Five commitments explain every choice in the surface that follows.

The surface at a glance

The seven tools

Tools are model-invoked. Each call is parameterized, executes a versioned Cypher pattern against the context graph, and returns a determination response.

release_trust(ref)

Answers whether a release ref conforms to its governing requirements. Returns conforms, does_not_conform, or indeterminate, with blockers and proof-chain references when applicable. The release-time gate, callable from CI directly. Composed by the idora-release-trust Skill.

file_context(file_path)

Returns the governing requirements, current verification status, and verification history for a file. Edit-time agents call this before modifying code. Reviewers call this when assessing a change. The response is a single projection: file metadata, governing requirements, applicable verifications, and proof-chain references. Composed by the idora-context Skill.

proof(receipt_id_or_claim)

Returns the full receipt chain for a determination. Given a receipt or a claim, returns the chronologically ordered chain of receipts, files, artifacts, evaluator identity, prompt versions, and code provenance where applicable. The audit primitive. Composed by the idora-proof Skill, and called as drill-down from release_trust blockers and file_context verifications.

gap_detection(scope)

Returns coverage gaps in a defined scope: requirements without conforming verification, files without governing requirements, unresolved contradictions in requirement relationships. Categorized output with entity references. Composed by the idora-coverage Skill.

drift(ref_a, ref_b)

Returns what changed between two refs: files whose content changed, verifications whose determination changed, requirements that gained or lost coverage, violations that emerged or resolved. The temporal-comparison primitive. Composed by the idora-drift Skill.

browse(query_shape)

The bounded exploration entry point. Returns paginated lists shaped per the query (requirements by source, files by repo, recent receipts by type). Each entity follows the canonical projection. Used directly. Not Skill-mediated in v1.

bulk_proof(refs[])

Assembles receipt chains for multiple subjects in a single call. Optimized for batch attestation: quarterly audit packages, regulatory framework conformance, periodic reporting. Composed by the idora-proof Skill for batch use cases. Called directly by compliance systems generating documentation.

The two resources

Resources are URI-addressable, read-only, and stable across calls within a version. They complement tools rather than compete with them.

pipeline_status

Exposes ingestion state: ingestion_lag_seconds, queue_depth, last_push_timestamp, push_count, days_active. Updated on every receipt write. Consumed by tools’ metadata blocks for graph-maturity context and by Skills for cold-start narration.

schema

Exposes the receipt and graph schemas at the envelope’s current version. Returns receipt schemas per receipt type, entity projection definitions, the canonical decision-shape vocabulary, and the current schema_version. Consumed by third-party Skill builders, by Skills declaring envelope-compatibility ranges, and by tooling generating client SDKs.

The metadata block

Every tool response carries a metadata block describing graph state at the moment of the call. Callers gate confidence on this metadata. Cold-start sparseness shows in the metadata, not by failing the call.

The block contains five named fields:

• ingestion_lag_seconds. How stale the graph is relative to the most recent push.
• graph_maturity. The substrate density signal: push_count and days_active.
• envelope_version. The API contract version.
• schema_version. The receipt and graph schema version.
• evaluator_model. The model identity for verifications surfaced in the response, where applicable.

A representative metadata block from a release_trust response:

{
  "ingestion_lag_seconds": 4,
  "graph_maturity": {
    "push_count": 847,
    "days_active": 142
  },
  "envelope_version": "1.0",
  "schema_version": "1.0",
  "evaluator_model": "claude-sonnet-4-20250514"
}

A release_trust response on a 7-day-old graph reads differently from one six months in. The determination is the same shape. The confidence the caller places in it is gated by the metadata.

The Skills layer

Skills are the consumption-layer surface, the path humans take to reach Idora through coding agents. Five Skills ship in v1, calibrated to platform engineer and compliance/audit early-adopter workflows.

Skills are SKILL.md packages following Anthropic’s open Agent Skills standard. They run in consumption-layer agents (Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Claude Desktop), not in Idora’s infrastructure. They are distributed via Anthropic’s Skills Directory and Vercel’s skills.sh.

Skills compose tool calls. They do not bypass the integrity boundary. A Skill cannot fabricate a determination, generate a receipt, or invent evidence. It recognizes the natural-language pattern, constructs the appropriate envelope call, and narrates the response. Cold-start framing and substrate sparseness are surfaced honestly through the metadata block.

The five Skills cover six of the seven tools. browse is intentionally outside v1 Skill scope; exploration is better served by ecosystem-level Skill discovery than by an Idora-authored navigation Skill.

CI pipelines bypass Skills and call tools directly with structured arguments. Same envelope, two consumption styles.

How teams and agents interact with the surface

Three patterns. Each shows the surface in motion.

Cold-start behavior

The surface is honest about substrate density. The graph compounds with every push. Early in a deployment it is sparse, and the surface reflects that.

A release_trust call against a 7-day-old graph with 12 pushes returns the same response shape as one against a 6-month-old graph: determination, confidence, blockers, metadata. The difference is in the metadata. Skills surface this explicitly to the caller: “Your graph has 12 pushes recorded; trust scores will become more reliable as more verifications accumulate.”

The surface does not fail on sparse substrate. It returns indeterminate where evidence is insufficient and surfaces the substrate state in metadata so callers act on it knowingly. Cold-start is a first-class condition, not an error.

What’s stable in v1

Integration is against stable contracts. The following are stable v1 commitments:

What is not fixed at the contract level: the specific Cypher patterns each tool executes against the graph, response-shape extensions per tool (additive, never breaking), and the receipt schema (versioned via schema_version and surfaced through the schema resource).

v1 is contract. Underneath, implementation evolves.

Closing

Seven tools. Two resources. Five Skills. A surface this size can be learned in an afternoon, called correctly the first time, and relied on permanently.

Readers wanting to understand the graph structure these tools traverse should read The Context Graph. Readers wanting to see how the surface evolves as integrity becomes cross-service should read Architecture Overview. Readers evaluating whether to pilot Idora should contact engineering@idora.dev directly. The TypeScript contracts, authentication details, and integration guides are shared during design-partner onboarding.