The ledger

GET /api/v1/repos/{name}/affected — topologically-sorted rebuild order after touching a shared crate. Reverse-dep walk (depth 3) sorted focus-first so each tier can be built once the prior tier is ready.

GET /api/v1/agents/:id — single agent detail by id, enriched with last_seen_at. 404 when the id is not in the registry.

GET /api/v1/agents — returns the full agent roster from data/agents.yml, enriched with last_seen_at (max last_commit_at across owned repos). Powers the /agents lab-inhabitants page on atlas-web.

GET /api/v1/agents/{id}/memory — live memory-directory summary for alice-persona-tycho. Returns file count, last-updated, and categorised title listing parsed from MEMORY.md. Other agents return a privacy-stub.

AgentRow — per-agent shape from data/agents.yml extended with model_id, role_tags, bio, and runtime-computed last_seen_at (max last_commit_at across scope_repos).

Axum AppState — wraps pg pool + HS256 JWT verifier (openalice-axum-common) + internal secret + TenantPools registry (Phase A). Cloned per request via Axum's State extractor.

Org task board — GET /api/v1/board (grouped Now/Queue/Blocked/Done view incl. live work-claims per agent), GET/POST /api/v1/board/tasks, PATCH /api/v1/board/tasks/{id}. MCP: atlas.board.list / atlas.board.add / atlas.board.update.

GET /api/v1/code/callees — forward AST call edges (what symbol X calls) over the code_calls index, optional repo filter. Returns callee_name/symbol/file + confidence.

GET /api/v1/code/callers — reverse AST call edges (who calls symbol X) over the code_calls index, optional repo filter. Returns caller_symbol/file + confidence.

GET /api/v1/code/symbols — substring search over the code_symbols index (optional repo + kind filters). Returns symbol name/kind/file/line so agents can locate a definition without grepping.

GET /api/v1/commit-sparks — batch per-repo 90-day daily commit counts from commit_history ([{repo, days:[{d,c}…]}]). One query, one response; powers the /repos catalogue sparklines.

GET /api/v1/context?repo=X — one-call session-orientation pack: repo shape + grade, 7d commits with subjects, open board tasks, active claims, drift counts, canon links. MCP: atlas.context.

Layered data-flow graph — GET /api/v1/repos/{name}/dataflow?zone= classifies symbols into 7 pipeline layers via path-glob rules, aggregates repo_file_graph + code_calls into weighted module→module edges, flags dead/layer-skip/back-edge anomalies, joins dup_score from Phase-1 clusters. Product-group keyed layer rules (alice-core ships now; others fall back to unclassified).

Returns full body_markdown + parsed frontmatter for a single lab digest (e.g. 2026-W21). 404 when label not found.

Lists all lab digest files from knowledge/digests/, sorted date DESC. Each item includes label, title, date, status, tags, authored_by, a 120-char body preview, and the first ≤3 top-level headline bullets parsed from the markdown body.

Org-wide planning-vs-code drift surface. Closes the goal⇄feature matcher loop by surfacing four classes of mismatch: 1. goals_without_features — live goals (planned / in_progress) with zero matched @feature annotations. Likely roadmap gaps. 2. features_without_goals — @feature annotations that no live goal in the same repo matches. Likely undocumented work. 3. stale_goals_done_but_no_features — goals marked `done` but no @feature backs them up. Either the doc lies, or the feature was never annotated. 4. stale_features_for_dropped_goals — features matched ONLY against goals the operator dropped. Likely dead code. All four reuse the exact `count_matched_features` matcher used by /api/v1/repos/{name}/goals so the cockpit's progress bars and this drift surface always agree on what counts as evidence.

Drift-surface: reads the structured `drift_events` table (written by the ingestor's doc-drift and stale-feature passes) and turns each open event into an *actionable item* an agent can act on directly. Two endpoints share one query helper: GET /api/v1/drift/open — org-wide, all repos GET /api/v1/repos/{name}/drift — per-repo slice Both return the same JSON shape. Query params: kind — filter to one kind (e.g. "doc-drift") min_severity_days — minimum severity_days threshold (default 0) Items are ordered by severity_days DESC (most overdue first) so the caller can cut off after N items or consume the full list. `suggested_action` is derived from `kind` + `target` so every item is immediately actionable without a second lookup.

Org-wide drift surface from drift_events — one actionable item per open event with suggested_action, ordered by severity_days DESC. Filters: kind, min_severity_days.

Per-repo actionable drift surface from drift_events — same shape as /api/v1/drift/open but scoped to one repo.

Org-wide planning-vs-code drift — 4 buckets (orphan goals, orphan features, stale done goals, dropped-goal features). Reuses count_matched_features so cockpit progress bars and this surface always agree on evidence.

Per-repo planning-vs-code drift — same 4 buckets as /api/v1/drift but scoped to one repo.

POST /api/v1/features — agent-authored feature into agent_features (separate from scanner-owned features). Agent-key auth + require_scope(repo, openalice). Stamps authored_by from the principal; surfaces in /features and /repos/{name}/features tagged source:"agent". 401 anonymous, 403 out-of-scope repo.

POST /api/v1/goals — agent registers an Atlas progress-goal (category/title/status/priority/target/repo) into agent_goals (separate from the scan-owned repos.manifest_goals JSONB, so a scan never wipes it). Agent-key auth + require_scope(repo|openalice, openalice). Stamps created_by from the principal — never the body. 401 anonymous, 403 out-of-scope.

GET /api/v1/repos/{name}/goals — manifest goals enriched with heuristic progress% derived from @feature token-overlap matching.

GET /api/v1/grades/regressions — cross-repo axes whose latest score dropped ≥min_drop below the prior snapshot. Surfaces quality regressions that UPSERT would silently overwrite.

POST /api/internal/grades — UPSERT per-(repo,axis) quality grades {repo, source, axes:[{axis,score,grade?,summary?}]}. The live write path for the recurring LLM architecture-grader; replaces the frozen 2026-06-10 SQL seed so /grades recomputes. Requires X-Internal-Secret.

GET /api/v1/repos/{name}/grade-trend — per-axis grade history from grade_history (newest first). Shows how each quality axis evolved over time. Optionally filter by axis and cap rows.

GET /api/v1/repos/{name}/hotspots + GET /api/v1/hotspots — repo-level CodeScene-style churn×debt hotspot ranking from commit_history + quality_breakdown.code_health. Honest limitation stated in response.

POST /api/v1/marketing/image-gen — Codex-hosted image_generation tool exposed as a stable HTTP endpoint with every documented parameter (size, quality, format, moderation, background, output_compression, input_fidelity, n, multi-reference edit mode, output_path). WAF-friendly user agent + cookie jar + SSE drain + base64 decode + optional disk write under marketing/ root.

GET /api/v1/repos/{name}/impact — reverse dependency walk returning every consumer of the focus repo up to configurable transitive depth. Edges from runtime_deps_manifest only; depth clamped 1–5.