/cass

name: cass
description: 'Search agent session logs and cited episodes with CASS. Use when: past prompts, decisions or failures may answer a question; repeated text is not a proven lesson.'
practices:
- pragmatic-programmer
skill_api_version: 1
user-invocable: true
hexagonal_role: supporting
consumes: []
produces: []
context_rel: []
output_contract: 'stdout: cited session hits (source_path + line) and index-state facts; no artifact directory unless an explicit -o export path is given'
metadata:
  dependencies: []
  capabilities: [cass]
  effects: [rebuild_local_index, sync_remote_sources, download_semantic_model]
  canonical_status: canonical
  disposition: keep_optional_adapter
  tier: execution
  external_dependencies:
  - "cass binary (>=0.3.6 recommended; some commands require HEAD \u2014 see Version Pinning)"
  - jq (required for parsing --json output)
  - GNU coreutils 'timeout' (recommended; cass index can hang under contention)
  - ssh + rsync (optional; only for cross-machine `cass sources` workflows)
  - fastembed model bundle ~90MB (optional; only for --mode semantic / hybrid; install via `cass models install`)

cass Session Search

> **Core Insight:** Recurring prompts identify candidates to investigate. Repetition can reflect success, repeated failure, copied instructions or retries; check the surrounding evidence before reuse.

`cass` is an upstream (Dicklesworthstone) tool and is **self-describing** — do not re-learn its surface from this skill. Discover it live:

cass capabilities --json      # features/connectors/limits of the installed binary
cass introspect --json        # full schema of every command + response
cass robot-docs guide|commands|examples|schemas|contracts   # machine-targeted docs

This skill carries only the AgentOps operating doctrine: when to reach for cass, the discovery workflow, the recovery posture, and the anti-patterns we have actually hit.

Constraints

• Never run bare `cass` because it launches a blocking TUI; use a JSON, robot, or explicit file-output command.
• Within authorized sources and destinations, treat a stale index as searchable

and refresh it only when the selected invocation permits indexing, using a bounded background command; stale is not broken.

• Preserve source sessions and require explicit permission for destructive cleanup; recovery may rebuild only derived index state.

Authorization and episode association

Before any search, view, context lookup, export, index recovery or sync, check source-owner, task, model/provider and destination authorization. Indexes, hit metadata, source paths and tracker comments inherit source restrictions; read permission does not permit forwarding to another model or versioning in Git. The recovery defaults below apply only inside this authorized envelope; source sync and model downloads also require the selected egress authority. Unavailable controls leave restricted-source operations unavailable, without retrieving first and redacting later. Missing, restricted, unavailable, no-match and insufficient evidence are different outcomes.

Capture work identity at dispatch/start and observed native IDs at startup through the caller-owned native comments/metadata or runtime facts, independently of handoff. Use [SESSION_FORMATS.md](references/SESSION_FORMATS.md#work-to-session-associations) for source-store/work identity, parent/resume provenance, supported work spans, permitted locators and available frozen source bounds/digests. Native logs retain execution authority; CASS discovers candidates. Search/view/expand/context output does not prove a complete episode was read or that a related hit is a parent. Record query, filters, limit, index freshness and discovery cutoff; zero hits means no match within those observed limits, not absence everywhere. Missing children, new tails and unknown source lengths remain explicit. T09 owns later coverage verification; this skill does not implement or certify it.

When to Use

• "What did I ask last time?" / "find that prompt that worked" — session archaeology
• Prior-art check before inventing a new approach, plan, or prompt
• Scope archaeology: "when did we decide NOT to do X?"
• Post-context-loss recovery: what was searched for after a crash = what mattered

Folded triggers (ag-s43tg wave 1): `casr` + `cass-memory` route here

• **`casr` → cross-harness resume.** [RESUME.md](references/RESUME.md)

distinguishes `cass resume` (same-harness command resolution) from the separate cross-harness converter. `cass context` discovers candidate relations; verify a native parent link before choosing a parent session to resume.

• **`cass-memory` → `cm` procedural memory.** When procedural-memory retrieval is relevant, use the caller-selected memory source. CASS can supply episode evidence for a concrete uncertainty; retrieval does not authorize promoting a lesson or changing instructions.

History as Evidence

Your conversation history contains:

• **Prompt revisions** — Compare what changed and what happened afterward
• **Recurring prompts** — Investigate repetition without assuming effectiveness
• **Scope decisions** — "When did we decide NOT to do X?"
• **Recovery moments** — What you searched for after context loss = what mattered

Native execution remains the default. Use CASS when a prior decision or episode could resolve a concrete uncertainty, or when the caller requests history. It supplies cited episodic evidence on demand, never policy; AgentOps maintains no merged corpus around it. Routine work needs no mining pre-step.

Bounded History Routing

When history is relevant, select a bounded search (for example one query family, `--fields minimal`, a real `--limit`, under a minute of wall clock). Three outcomes have different implications:

• **Direct hit** — a prior session addresses this. Inspect the user intent,

action, observed result and corrections before deciding whether to reuse it; cite `source_path` a