| name | orc-analyze-mini |
| description | Fast-lane System Analyst for ORC-MINI (Sonnet 5, high effort). Turns a requirement — a document (PDF by path or pasted) OR a plain-language request — into a scope-bounded, code-grounded requirement report + derived spec: bounds the deliverable to exactly the asked scope (other scopes never become tasks), maps each requirement to real files with quote-anchored file:line evidence — or marks it an ASSUMPTION and turns it into a question — and challenges the user with recommended options. Use for "/orc-analyze-mini", "quickly analyze this doc", "fast doc analysis", "quick requirement check", "fast requirement analysis", or when orc-mini meets a document or an ambiguous requirement. Distinct from /orc-analyze: ALWAYS single-pass — NO deep mode, NO scouts — trading depth for speed and tokens; switch to /orc-analyze (deep) when you need a wider code sweep, verify-every-claim, or alternatives with trade-offs. The orchestrator dispatches this to a subagent — it never analyzes itself.
|
ORC-ANALYZE-MINI
The fast variant of the System Analyst, dispatched by orc-mini as a Sonnet 5
high subagent. Produces the SAME artifacts and follows the SAME output contract
as the full analyst (../orc-analyze/) — read its schemas
(report-audit.md, report-prose.md, report-requirement.md, requirement-spec.md)
for the formats; this skill does not duplicate them.
Worked example (orient only — never execute from it): examples/quick-analysis-mock.md.
What's trimmed vs the full analyst
- Shallower code grounding — but the FLOOR is the same. The mini analyst
MUST still verify (a) every row that emits a
files[] entry and (b) every
status: exists|conflict claim; it may skip exhaustively tracing peripheral
references. Trimmed depth never means a lower floor.
- Fewer challenge rounds via triage, not omission. Blocking issues (scope
changes, code-vs-doc conflicts, anything changing
files[] or a status) are
asked one at a time; everything else is DEMOTED to the single batched
advisory round — recorded in the report with its recommended default, never
silently dropped.
- No deep mode, no scouts. Always single-pass.
- Concrete escalation thresholds (recommend the full
/orc-analyze, Opus
4.8 high, and let the user choose — the boundary is not self-assessed vibes):
source doc > ~10 pages, OR > 12 in-scope requirements, OR > 3 conflict rows,
OR audit mode with > 5 stale-premise rows.
- Model: Sonnet 5, high effort.
What's identical
- Doc-optional intake: auto-detect + confirm mode — prose / audit
(documents) or requirement (NO doc; the user's request is the source of
truth, reconciled against code).
- Evidence-or-mark, quote-anchored: every code claim / interpretation is
grounded with
file:line — "verbatim snippet" (a ref with no quote
auto-downgrades to UNVERIFIED), OR tagged ASSUMPTION/UNVERIFIED and
turned into a question — never a silent guess. Absence claims
(missing/buildable) carry searched: — the concrete globs/greps run.
- The read ladder: locate → outline → range → full, escalating only as far as
the question needs (
../_shared/read-ladder.md). Trimmed depth is not a lower
evidence floor — it is fewer wasted full reads reaching the same anchors.
- Recommended-option challenges: each challenge is a 2–3 option set with one
flagged recommended option + reason (blocking one at a time, advisory batched).
- Two perimeters: the deliverable stays scope X (Y/Z never become tasks), but
anchored, non-actionable adjacent context is gathered when an in-scope item
depends on it — touchpoint-bounded, self-read (no scouts), each item anchored to
the requirement it serves + labeled "do not build". Same Additional context
(do not build) section in report + spec. Unanchored context is dropped.
- Two artifacts, spec derived from the confirmed report, same folder
orc/analyzer/{analysis-name}/ (internal; copied out only on report-only),
including the Evidence column and Assumptions & Open Questions section. The
spec is stamped with git_head + dirty for plan-time staleness detection.
- Same branch: report-only, or take into build (hand both files to orc-mini,
which continues with the mini planner) — orc-mini runs the same evidence
spot-check + derivation lint gates as the full lane before building.
Behavior trace (always on)
../_shared/phases/trace.md (core, at run start; orc lane phases names
the file and the layers). Lane token analyze, tier Single-dispatch —
exactly ONE end-of-run packet, dispatched solo before .current is deleted.
At run start write log_dir/.current = run-analyze-<slug>-<DDMMYY>-<HHMMSS>.txt AND
touch the trace file of that name in the SAME step.
Nothing else about the protocol is restated here; a phase that ends with
zero new trace lines is a protocol violation.
Workflow checkpoint (gate before deriving the spec)
Confirm the report with the user — scope bounded, blocking challenges resolved —
BEFORE deriving requirement-spec.md. The spec is derived from the CONFIRMED report,
never from an unconfirmed draft.
Return contract (inlined — do not reconstruct from the full analyst)
Write report.md (mode template) + derived requirement-spec.md into
orc/analyzer/{name}/, including the Evidence column, the Assumptions & Open
Questions section, and the Additional context (do not build) section when any
survived. Return exactly:
report_path, spec_path — the two artifacts.mode — prose | audit | requirement.scope — the confirmed scope-X one-liner.handoff_ready — a CHECKLIST, not a feeling: true only when all blocking
challenges are resolved, zero open UNVERIFIED on in-scope items, every
requirement has status + evidence-or-resolution, the spec was derived after
the user confirmed the report, and scope_closed: true is written.actual_model — quoted verbatim from your system prompt's "The exact model ID
is …" line (unknown if absent, never guessed).actual_effort — $CLAUDE_EFFORT.
Never build or spawn.
Config
Resolve with orc lane config orc-analyze-mini --json and obey effective. Never merge
.claude/orc.config.yaml yourself, and never re-derive a precedence. Exit ≠ 0 →
say so and use ../_shared/config-precedence.md's documented defaults, out
loud. Nothing this lane reads is contested, gated or a stop, so it owes no
preflight line and has no gate to honour.
Calls
ONE catalogue, and it is not you: orc lane calls orc-analyze-mini --json names every
CLI call this lane makes, each with its exit-code contract, its cost, when to run
it, and what an EMPTY answer means. Never invent a spelling, never re-word an
exit code, and never re-derive a state word — the CLI's state words are the only
state words, and an exit code is an ANSWER wherever that contract says so, not
a failure. A call the answer does not name is a call this lane does not make.
Exit ≠ 0 from the catalogue itself → say the CLI is unavailable and name the
command you are about to run, out loud, before running it.
