Intelligence

The Intelligence tab (/repos/<owner>/<name>/intelligence) displays the curated intelligence store Colony builds for each repository. As agents analyze, plan, and implement issues, they extract and record structured observations — architecture patterns, invariants, failure modes, and operator preferences — and surface them here with evidence, quality scores, and lifecycle status.

Intelligence items accumulate passively over time. The more issues Colony processes for a repository, the richer and more reliable the store becomes. Approved items represent Colony’s working model of the codebase; they inform how agents approach future analysis and implementation work for this repository.

Intelligence kinds

Each item in the store has one of ten kinds, corresponding to the category of knowledge it captures.

Kind	Description
Architecture	Structural decisions and patterns that govern how the codebase is organized — layers, module boundaries, and key conventions.
Invariant	Constraints that must remain true across the codebase — data integrity rules, API contracts, or behavioral guarantees.
Workflow playbook	Established step-by-step procedures Colony follows for common task types in this repository.
Test strategy	Preferred approaches to testing and coverage — frameworks, patterns, and what constitutes sufficient test quality.
Failure pattern	Recurring error signatures and failure modes Colony has observed while implementing or reviewing changes.
Coupling	Identified dependencies and co-change relationships between modules or files that affect how safely changes can be scoped.
Operator preference	Instructions or preferences contributed directly by a human operator via the Add knowledge action.
Design decision	Recorded choices about implementation approach, library selection, or trade-offs made for this codebase.
Risk area	Parts of the codebase with historically high defect density or change risk that warrant extra care.
Implementation note	Contextual notes about specific implementations — gotchas, non-obvious constraints, or explanatory context for future agents.

Status lifecycle

Every intelligence item moves through a status lifecycle as Colony gains confidence in it.

Status	Meaning
Candidate	Preliminary item surfaced automatically; not yet verified or promoted.
Observed	Confirmed by multiple independent observations; more reliable than candidate.
Proposed	Submitted by an operator or pipeline process, awaiting review and curation.
Approved	Confirmed and active. Approved items are part of the curated store agents draw on when working issues in this repository.
Dismissed	Reviewed and rejected. Kept for audit purposes but not factored into future work.
Superseded	Replaced by a newer item. Preserved for historical context.

Candidate items may be promoted into the intelligence store as a result of retrospective analysis — see Retrospectives for how completed-issue lessons feed back into accumulated knowledge.

Note

Operator-authored notes (submitted via Add knowledge) land at proposed status. They are never auto-approved — a pipeline curation step is required before they reach approved.

Quality score and evidence

The list view shows a quality score (0–100%) rendered as a filled bar next to each item. The score is a float (0–1) that reflects how well-supported the item is by observed evidence. A higher score indicates stronger corroboration across multiple sources.

Click any row to expand it and load additional detail:

• Summary and details — the item’s prose description and any extended notes.
• Source — which pipeline component generated the item, and the agent name if applicable.
• Observation count — how many times agents have independently observed this pattern or fact.
• First / last observed — the date range of observations backing this item.
• Scopes — one or more scope kind / scope ref pairs that narrow the item to a specific file, route, module, or other code unit.
• Evidence — individual evidence entries (evidence type, optional ref or path, and a link if a URL is available). Evidence entries load on row expansion via GET /pipeline/intelligence/:id.

Filtering and pagination

Three filters narrow the intelligence list:

Filter	What it does
Status	Show only items at the selected status (candidate, observed, proposed, approved, dismissed, or superseded).
Kind	Show only items of the selected kind.
Scope kind	Show only items whose scope set includes an entry with the typed scope kind (e.g. file, route, module).

The list is paginated at 25 items per page. Use Previous and Next to navigate.

Adding knowledge

Operators can contribute intelligence items directly from the issue action drawer. Open any issue in the pipeline (via Pipeline or Inbox) and click Add knowledge in the action panel.

The Add knowledge modal accepts:

Field	Required	Description
Title	Yes	Short label for the note (200 characters max).
Summary	Yes	One-line description of the knowledge being recorded.
Details	No	Optional extended explanation.

Submitting posts to POST /pipeline/repos/:owner/:repo/knowledge. The item lands at proposed status with kind operator_preference and the issue recorded as provenance. It appears in the Intelligence tab immediately and follows the normal curation lifecycle before reaching approved.