Step 2 — Resume Session

1. Read state

Read .doc-this/state.json and .doc-this/plan.md.

1a. Plugin-version skew check (advisory)

Read the installed plugin version from ${CLAUDE_PLUGIN_ROOT}/.claude-plugin/plugin.json (.version field). Compare with state.json.plugin_version:

• If equal, or if state.json.plugin_version is null (legacy state file): proceed silently.
• If different: print one line — "⚠️ Doc-This: state was created on plugin v<old>, currently running v<new>. Resume continues; rerun a phase with --regenerate=<phase> if you want it re-done under the new version." — then proceed. Do NOT block.

After the check, write the current plugin version into state.json.plugin_version so subsequent resumes compare against the up-to-date stamp.

1b. Structural extraction re-check

If state.json.structural_extraction exists:

• LSP: Re-run ToolSearch("select:LSP") to ensure the deferred tool is loaded in this session. Run a quick documentSymbol on a known file to verify LSP is still responsive. Update lsp_available if it changed.
• UA staleness: If ua_detected is true, compare ua_commit_hash with current git rev-parse HEAD. Update ua_staleness flag. If UA was stale before and now matches HEAD (user re-ran /understand), clear the staleness flag.

If structural_extraction does not exist (legacy state from before this feature), run the full check from step-01-first-run.md step 4a.

2. Greeting

Say: "[Name], welcome back to Doc-This."

3. Progress summary

Show:

• ✅ Completed phases (completed field)
• 🔄 Current phase (phase field) with the last checkpoint recorded
• ⏳ Pending phases (pending field)

Example:

"Progress so far: ✅ Reconnaissance done 🔄 Analysis in progress — modules billing and invoicing analyzed; payments and reports pending. Coverage: 1720/2000 source files; 280 pending, resuming at src/payments/refund_form.aspx ⏳ Interpretation, Synthesis, Generation, Review"

3a. Coverage cursor

If state.json.coverage exists and files_pending > 0: resume the Code Analyst from coverage.cursor (module + next_file) — the ledger (.doc-this/context/coverage-ledger.json) is append-only, so the unread set is always manifest source ∖ ledger. Include the coverage line in the progress summary (above).

If coverage is absent and the run is past reconnaissance, this is a legacy (pre-coverage) run — mention once: "This run predates Total Source Coverage; /doc-this --backfill-coverage brings it into compliance."

3b. Legacy runnability check

If legacy_runnable is absent or null (legacy state from before this field), ask the step-01 §2a question once and persist the answer. Never re-ask when set.

4. Database context check

If database_ownership is null and Scout has completed (Scout is always before step-04), prompt to run step-04 before continuing. Otherwise carry on.

5. Gap-answer mode

If answer_mode = "file":

"Reminder: your answers to the gap questions go in .doc-this-sdd/questions.md. Let me know when you're done."

If answer_mode = "chat" (default), no extra prompt.

6. Confirmation

Ask only: "Continue from where we left off? (CONTINUE to proceed)"

After confirmation, resume the next pending task in .doc-this/plan.md.

🚫 Do NOT offer /clear + /doc-this here. The user just resumed a fresh session; suggesting they clear and reopen is redundant. The preventive-pause prompt (described in SKILL.md under "Preventive pause between heavy steps") only fires after an agent completes work in this session — never on the resume greeting itself.

See references/checkpoint-guide.md for state.json write rules.