| name | recon |
| description | Ground a LCD work-item in CURRENT best practice before deciding on an approach. Resolves up-to-date library/API/tooling guidance via Context7 and a targeted web search, in a subagent, and returns a few dated bullets. Use at the start of Standard/Deep work that touches a library, framework, SDK, API, or pattern whose current best practice you're not certain of — before writing the plan or choosing a dependency. |
| user-invocable | true |
LCD Recon
Answers "what is the current best way to do X?" before the architecture is committed — so the plan
isn't anchored to a stale training cutoff.
When to run
- At the start of Standard/Deep work that involves a library/framework/SDK/API/CLI/cloud
service, a new dependency choice, or a pattern you're not current on. Skip for Quick and skip
when the work is plainly within well-known, stable territory (don't burn tokens confirming the
obvious).
How to run (lean-orchestrator rule)
Run recon in a subagent so the token-heavy doc/search dumps never enter the orchestrator's
window. The orchestrator receives only the digest and writes it to the work-item JOURNAL.
Dispatch the plugin's lcd-recon agent (agents/lcd-recon.md) — it is read-only with the
research tools and the method baked in (installed-version check → Context7 → targeted web
search), and its definition carries the digest contract: { findings: 3–6 dated bullets, avoid: [...], recommendation: one line }, ≤15 lines total. The dispatch prompt therefore only
needs the work-item context: slug, the library/API/pattern in question, and the installed
version if already known.
When recon is dispatched inside a Workflow (e.g. the Deep-lane orchestration), pass that
digest shape as the agent's schema so the contract is validated at the tool layer rather
than hoped for.
What the orchestrator does with the result
- Append the bullets to the work-item JOURNAL LOG (dated).
- If a finding drives a real architecture/tooling choice, record it in
<root>/DECISIONS.md
(cite the recon finding as Context) and reference it from plan.md's Architecture section.
- Proceed to plan/build with the current-practice grounding in hand.
Keep it light
Recon is a check, not a research project. One subagent pass, a handful of dated bullets, a decision.
If the answer is "the approach you already had is still current," that's a valid one-line result —
record it and move on.