메뉴
(new) Find a person's repeated tasks from their screen activity, and what each one is worth automating, with the time and money saved. Outputs the numbers and data a report is built from, not the visual. Use to analyze processes, mine workflows, or see what is automatable and what it saves. Every figure is labelled and grounded, never made up.
Write step-by-step automation instructions for a workflow, tailored to your tool (Claude, n8n or Zapier). Best run right after discover-patterns or process-analyst.
(new) Turn a process analysis into a polished, client-ready report, first an HTML deck you review, then a matching PDF. Use to package an analysis into an exec report or PDF. Shows the deck for approval before making the PDF, and never makes up numbers.
Generate a time report grouped by client/project
Discover repeated workflow patterns from screen activity and suggest automations
Summarize what you've been doing in the last 30 minutes.
Prepare a MemoryLane release by updating the version and release notes, then creating and pushing the tagged release commit that triggers CI. Use when the user asks to release, ship, publish, bump version, or cut a stable or prerelease version.
| name | process-analyst-new |
| allowed-tools | mcp__memorylane__browse_timeline, mcp__memorylane__search_context, mcp__memorylane__get_activity_details, mcp__memorylane__get_user_context, mcp__memorylane__list_patterns, mcp__memorylane__get_pattern_details, Read, Write, Task, Bash, AskUserQuestion |
| description | (new) Find a person's repeated tasks from their screen activity, and what each one is worth automating, with the time and money saved. Outputs the numbers and data a report is built from, not the visual. Use to analyze processes, mine workflows, or see what is automatable and what it saves. Every figure is labelled and grounded, never made up. |
Turns raw screen-activity rows into a defensible breakdown of a person's repeated work, what it costs, and what to automate. A separate packaging skill renders the visual. Your job is truth and numbers, not persuasion. Bar = 80/20: directionally accurate, grounded, never fabricated. Can't ground a claim? Weaken its label or drop it.
Verified against the MemoryLane MCP + storage code; everything below references these:
browse_timeline/search_context) return start only; end is stripped. Only get_activity_details gives a real [start -> end]. Timeline duration = gap approximation (next start minus this, mixes idle) = reasoned, never seen.browse_timeline defaults to sampling="uniform" + a row cap; header reads Showing X of Y when sampled, else N activities. Sampled counts = lower bound. search_context no-query = recent_first (drops oldest). Never count/sum from a sampled pull.window_title (PR/ticket #, doc) = seen; invoice/order IDs live in OCR (get_activity_details), sparse = reasoned.search_context only ranks a query vs the corpus.tld/OCR not in the timeline (tld only via get_activity_details).endTimestamp+durationMs to the timeline, a get_activity_stats aggregate tool, and tld to the timeline (primitives already exist).Intake gate: if the prompt omits the context below, ask for all missing fields in ONE batched AskUserQuestion first (offer defaults). If skipped, proceed with labelled estimates. Never silently assume.
derivation = base rate + source + loading multiplier (e.g. x1.3). Never show a $ figure without source + multiplier.tld/OCR; durations here are seen, not reasoned. Read via SQL (Bash); the upload mount is read-only, so copy the file to a writable folder before querying. (2) Export (CSV/JSON) with end times + url/tld. (3) Live MCP = start-only (durations reasoned), samples (counts lower-bound), no tld. More direct data = more "seen" + tighter numbers. State source + tier in meta.meta. State the window used.Preferred: one sub-agent per stage via Task + a reviewer, passing structured output forward. Fallback (no sub-agents): run stages in order yourself. Gates are mandatory. A human review gate (Checkpoint, after process discovery) pauses the run: detailed quantification + deliverables run only after the user approves the identified processes and direction.
Stage 0 — Ground & scope. Confirm inputs; resolve cost basis (rate + derivation + label). Pull get_user_context (prior, not truth). Record coverage + blind spots (per Data limits). Gate: cost basis has formula/source/label; coverage + n=1 scope stated.
Stage 0.5 — Descriptive baseline (credibility floor). Anchors the report; supplies the denominator/ceiling even with zero processes found.
tld/title-host to a domain in code, then map to a clean name via a lookup table (mail.google.com -> Gmail, *.atlassian.net -> Jira, ...); AI only names misses. Non-browser apps use app_name. Shows Jira/Gmail/Notion, not one "Chrome" bucket.data_coverage_notes. Nothing runs on an unreviewed list.N activities; split sub-ranges if needed). Never compute a share from a sampled batch.coverage = total_active_hours / (work_hours_per_day x active_days) (default 8 h/day). If coverage < ~40%, enter sample mode: label everything "from a captured sample of Xh (~Y% of work time)"; still give the capacity/savings headline but as a clearly-labelled estimate (sample-based, wide low/base/high range, coverage caveat, confidence Low), and prefer expressing capacity as "% of the work we observed" before projecting to the full role/team; force overall confidence to Low; put "record more continuously (coverage too low)" at the top of what_would_sharpen. Coverage is a lower bound on representativeness, not proof the sample is unbiased.
Gate: % per app sums ~100; contributing days unsampled (or from file); active-days exact; coverage computed and sample mode set if low.Stage 1 — Sessionize. Pull day-by-day, de-sampled. Segment by priority: (1) case-hints (primary) from window_title (seen); if titles weak, a bounded OCR sweep (get_activity_details, <=100 ids/call) for IDs (reasoned, require corroboration, never join on a bare number); (2) topic / content continuity from summary text and (when titles are weak) the OCR content, used to set run boundaries, not only to find IDs, so two runs that share the same app-set (e.g. two different Sheets+Slack sessions) are split by what the content is about, not merged because the apps match; (3) app-sequence; (4) time gap (last-resort tie-breaker only), 5-min default, calibrated to this user's start-delta distribution (state value + why). The gap is the weakest signal: never let it be the primary boundary when content/topic can separate two runs. Pre-merge adjacent rows sharing app+hint (and rows split only by the 5-min cap or an app change) into one step before any gap rule. Record the ID-corroborated share (K of N runs carry a stable cross-app ID) rather than applying a hard cutoff: label frequency "reasoned, ID-corroborated on K of N runs", and let confidence scale with that share (graded, §4), not a binary cliff. When K/N is low, lean harder on content continuity (above) before the gap, and say so. Output = ordered activity-level steps (start/stop, hint, apps, gap-duration reasoned, idcorroborated bool); reasoning log for non-obvious boundaries. _Gate: no sampled day; no single-row split artifact; durations reasoned; ID-corroborated share recorded.
Stage 2 — De-pollute. Flag interruptions (Slack/email peeks, personal messaging, browsing, entertainment): excise from task time but log them; keep only if OCR/summary shows the content fed back in. Remove pure-discard sessions: personal messaging; learning/studying; general browsing/social/news/shopping; programming + code review; entertainment; generic inbox/Slack triage unless it triggers a cross-app workflow; standalone IDE/file management. Reasoning log per excised chunk. Gate: kept interruptions have an evidence reason; discards match the list.
Stage 3 — Cluster into repeated processes. Cluster by hint-type + step-sequence similarity + app set. An instance = one reconstructed case; frequency = count of cases, not rows (log "N rows -> M cases"). A process repeats, is stable, and is operational. Cadence gate: >=3 cases in window, OR >=2 with a period the window can contain; for weekly/monthly hints (month-end, payroll, board pack) widen to 30 to 60 days via targeted search_context before deciding/annualizing. Below the gate, demote, do not drop. A cluster that looks recurring but misses the gate (1 to 2 sightings, or a likely weekly/monthly process the window was too short to confirm) is NOT discarded: tag it verdict: candidate (low confidence) and carry it into the Checkpoint table as a candidate-tier row, with its uncertainty named (e.g. "seen 2x in 21 days, possibly monthly, window too short to confirm"). Only true one-offs and creative/judgment work are excluded outright. This protects the high-ROI low-frequency tail (close, board pack, quarterly rollups) that a short window or strict count would otherwise hide. Per process: frequency, time-per-instance distribution (n, median, IQR/p90), total time, varies/constant, role/department. If estimated time-per-instance exceeds captured hands-on time, state the ratio + why ("captured 100 min, estimated 5 h, x3 for fragmentation"). Reasoning log: why each cluster is a process, a candidate, or excluded (one-offs + creative logged as considered-and-excluded). Gate: frequency traces to counted cases or a stated inference; time carries n + spread, reasoned.
Stage 3.5 — Common path (80/20). Describe the single dominant way it usually goes (~80/90% sequence) + rough coverage. Do NOT over-model (no variant trees, loops, exception enumeration). Note exceptions exist but aren't modelled, and that building the automation needs deeper hands-on analysis. Common path = automatable core; remainder stays human. Path + coverage = estimates. Gate: coverage + case-n recorded; output says it's an estimate needing deeper analysis to build.
Checkpoint, confirm processes + pick what to detail (human review gate, triage menu). Before the detailed half, STOP. First run a light provisional pass over EVERY qualifying process AND every demoted candidate (Stage 3), none dropped: rough automatable % (from category + work-nature, judgment cap applies), difficulty (Easy/Med/Hard), and uplift (provisional capacity-unlock % + net annual savings range), reusing the Stage 3 frequency + time and the cost basis. Everything here is provisional, refined by Stages 5, 5.5, 6 after selection. Then present one table listing ALL of them, confirmed processes first sorted by impact (annual hours) then occurrence (frequency), candidate-tier rows below:
| Pattern | What it is (1 sentence) | Time/yr (h) | Est. automatable % | Difficulty | Uplift (capacity % + net $/yr) | Tier + confidence |
The last column marks each row process or candidate (low confidence) with a one-line reason for any candidate (e.g. "possibly monthly, window too short"). Under the table, list only the true exclusions and why (genuine one-offs, creative/judgment work); sub-threshold-but-recurring work appears IN the table as a candidate row, never demoted to a footnote. Then ask the user (AskUserQuestion, multi-select the rows) which processes to dig deeper into, and allow drop / merge / re-scope / redirect. Run Stages 4 to 6 ONLY on the selected processes; the rest stay in the table and in the ledger as logged-but-not-detailed (selected_for_detail: false), so nothing found is hidden. Honor their edits. If there are more candidates than the question can list, show the table and ask them to name their picks. If no human is available (batch run), record "checkpoint auto-passed" in review.checkpoint, dig into the top processes by provisional ROI, and continue.
Stage 4 — Decompose steps (activity level, map-ready). Ordered steps: action (verb+object), app, input, output/artifact, time (reasoned), pain (reasoned), automatable (bool), note (what changes / watch-out). A row is not a step: derive steps from transitions across rows; don't split one row into separately-timed steps. Also capture for the map: 0 to 3 key decision points with branches (estimated, light, not variant trees) and a per-app per-run time rollup (sum step time by app). Steps + decisions + per-app times = the map-ready representation; a downstream map skill reads this and draws the visual (like the /patterns Map view), not here. Gate: each step maps to >=1 activity (cite IDs) or a labelled reasoned bridge; time/pain reasoned; decisions estimated.
Stage 5 — Map automation (decision tree). Classify into a category (Section 3). Bottom-up fraction: automatable_fraction = sum(step.time where automatable) / sum(step.time), re-derivable; map the 1-5 score to a band (5 -> 0.8-0.95, 3 -> 0.4-0.6, 1 -> 0.05-0.2); express as a range; name human steps in human_remainder. Judgment cap: Content Generation + Research & Synthesis fraction <=0.6 unless OCR shows boilerplate.
Verdict + route (walk every process through this tree, never forced):
effort_tier.Connector capability check (before recommending a route): for every API/MCP connector or plugin in the recommendation, fetch its latest official docs and confirm the specific read/write it needs; record server, read, write, how verified, and date in connector_checks. Confirm the exact product/edition the user runs (variants differ, e.g. Looker vs Looker Studio); if the surface tool cannot connect, pivot the route to the underlying data source and record that pivot in connector_checks. An unverified or write-incapable connector cannot back a Human-in-the-loop or Automated verdict (see Future state).
Future state (score every kept process on this ordered scale): No change (manual) -> AI-augmented (human operates, AI drafts/suggests) -> Human-in-the-loop (automation runs, human approves/handles exceptions) -> Automated (hands-off). Kills = eliminated (off-scale). Augmented vs HITL = who operates (human invoking AI = augmented; automation running with a human gate = HITL). Roughly MECE on one axis (how much the machine does); the tree picks the route, this labels the end state. Ordered, so a downstream report can render it as a chevron (highlight the step, grey the rest). Capability ceiling: the label follows verified connector capability, not assumption. Reaching Human-in-the-loop or Automated (the machine writes) requires a verified write to the target tool; if only read is verified (or nothing), cap at AI-augmented (AI drafts, human writes). E.g. sprint estimation only moves from "AI drafts" to "runs with a human gate" once Jira + Notion writes are confirmed.
Tie every recommendation to the manual step(s) it removes (cite step numbers); prefer an integration that eliminates the action over one that merely assists (e.g. a Cowork Figma plugin that fills the slide template from a CSV/JSON, removing the copy/paste re-keying, not "paste into Claude then drop it in by hand").
Name + frame honestly: name the process after the automatable slice + verdict, not the creative whole; creative/judgment work is augment ("AI-assist: structure findings + draft deck text; designer keeps the insight and builds the deck"), never "automate the slide deck". A priority label must reflect net ROI + verdict, never the gross size of a judgment-capped activity. Gate: tools exist in context or are explicit defaults; fraction re-derives; every process has a verdict, a route (if automate), a future-state label, the manual steps removed, and an honest name.
Stage 5.5 — Automation feasibility. Per candidate:
Stage 6 — Quantify savings + net ROI (ranges). All arithmetic in code.
annual_hours = time_per_instance x frequency_per_year; automatable_hours = annual_hours x automatable_fraction; gross_$ = automatable_hours x rate. Show inputs + labels.hours_freed = time_per_instance x frequency_per_year x automatable_fraction and annual_savings = hours_freed x loaded_rate. Never display one automatable % while deriving the dollars from a different effective fraction. A skeptic multiplies the printed inputs and must land on the printed savings; per-process savings must sum to the role/function subtotal and then to the combined total. Carry the exact inputs so a downstream page reconciles without re-deriving (see reconciliation in §5). Worked example (real build): slide-deck 3h/run x 2/month x 46% x A$105/hr -> ~A$3,500/yr; sprint-estimation 45m x 2/month x 28% x A$105/hr -> ~A$500/yr; Product subtotal A$4k. Marketing perf-reporting 27m daily x 90% -> ~A$6,200, SEM 35m x 8/month x 71% -> ~A$2,800, optimisation-logging 60m x 4/month x 60% -> ~A$2,000, all x A$70/hr; subtotal A$11k. Measured total A$15k; capacity 1h + 3h = 4h/week (5%); 2 + 3 = 5 processes.net_annual = gross_annual - annual_automation_cost; payback_months = build_cost / monthly_net. If payback > ~6 to 12 months, tool cost > savings, or high effort for a tiny saving, the verdict is leave as-is or redesign, not automate. Net, not gross, drives the verdict + ranking.activity_id -> process map; each activity in one process.list_patterns exists, cross-check frequency vs sighting counts and time vs durationEstimateMin (pre-counted, sampling-immune); cite as support, never add to your numbers; report discrepancies. Detector looks back ~1 day, so absent != nonexistent.capacity_unlocked_pct = annual saved hours / annual work capacity (this person; or the team if headcount supplied); pair with net $ saved. Headline reads like "automating these unlocks cost_basis.derivation states base salary, loading multiplier, work-year hours, and the resulting loaded rate, in this exact shape: Product "A$160k base -> A$105/hr loaded (x1.3 on-cost over ~1,950 h)", Marketing "A$105k base -> A$70/hr loaded (same basis)". When several ledgers combine downstream, each role/function keeps its own loaded rate; per-process savings sum to that function's subtotal, function subtotals sum to the combined total, and the exec headline restates the combined total. Never blend two functions onto one rate.
Gate (triple-test): re-derive top-3 savings independently; dedup map has no duplicate IDs; aggregate = sum of parts and sits under total_active_hours; FTE checked; inputs labelled; ranges present.Stage 7 — Systems landscape. Synthesis, last because it needs automation status. Inventory + time share come from the cleaned baseline (0.5); this stage overlays function grouping (Source/ERP, Comms, Docs, BI, Finance, CRM) + status (manual/partial/automated) from how each system appears in processes. Gate: every system appears in the cleaned baseline list.
Final Review — adversarial. Mechanical checks as code (aggregate = sum of parts; no duplicate IDs; % per app ~100; every headline has a range + label). Then break it: any number without label/evidence/range? sampled count? weak reasoning log? duplicate ID? team/FTE without headcount? automatable=true on judgment the data can't show is mechanical? tool not in context? aggregate above ceiling? click-level/fabricated detail? overall future-state label inconsistent with its steps (an AI-augmented or human-in-the-loop process with no human/judgment step, or an all-automate step list under an augmented label)? Slop sweep: every name/app/tool/step/number traces to a row or labelled assumption, else strip; deterministic numbers re-run identically. Threshold-sensitivity: re-segment one day at 3-min and 10-min gaps; if process count or total time shifts >25%, flag it. Write what_would_sharpen (2 to 4 cheap next steps that most raise confidence). Only survivors ship; record verdict + demoted/stripped.
RPA-era (deterministic): Data Shuttle (move structured data between apps), Reporting Ritual (same sequence on a schedule), Review Pipeline (queue -> cross-reference -> decide), Data Entry (read source -> type into forms), Alert Response (notification -> act -> return). AI-era (LLMs automate what RPA discarded): Content Generation (draft docs/emails/decks from inputs) and Research & Synthesis (gather across sources -> brief), both fraction <=0.6 unless templated. Use exactly these seven; if none fit, it's likely not an automation candidate (log why).
Impact x Automatability, Effort as tie-breaker; surface quick wins (high automatability + Easy) and big bets (high impact + Hard).Write two files, then a 5-line chat summary (don't dump raw output): process-analysis.md (readable) + process-analysis.json (for packaging). These filenames are the hand-off contract the packaging (report-new) skill reads; keep them.
{
"meta": {
"user_role": "",
"department": "",
"location": "",
"measurement_scope": "single observed user (n=1)",
"observed_user_count": 1,
"data_source": "",
"data_quality_tier": "",
"window": "",
"activities_analyzed": 0,
"active_days": 0,
"work_hours_per_day": 8,
"capture_coverage_pct": 0,
"sample_mode": false,
"cost_basis": { "hourly_rate": 0, "currency": "", "basis": "estimated", "derivation": "" },
"data_coverage_notes": "",
"blind_spots": "",
"threshold_sensitive": false
},
"baseline": {
"total_active_hours": { "value": 0, "basis": "reasoned" },
"active_days": { "value": 0, "basis": "seen" },
"app_time_share": [{ "app": "", "pct": 0, "hours": 0, "basis": "reasoned" }],
"deep_work_share": { "value": 0, "basis": "reasoned" }
},
"exec_rollup": {
"total_repetitive_hours_per_week": {
"low": 0,
"base": 0,
"high": 0,
"basis": "reasoned",
"evidence": []
},
"automatable_pct": { "low": 0, "base": 0, "high": 0, "basis": "estimated" },
"capacity_unlocked_pct": { "low": 0, "base": 0, "high": 0, "basis": "estimated" },
"hours_saved_per_year": {
"low": 0,
"base": 0,
"high": 0,
"basis": "estimated",
"error_sources": []
},
"dollars_saved_per_year_net": { "low": 0, "base": 0, "high": 0, "basis": "estimated" },
"fte_capacity_unlocked_this_person": { "low": 0, "base": 0, "high": 0, "basis": "estimated" },
"team_extrapolation": {
"headcount_supplied": false,
"value": null,
"basis": "estimated",
"multiplier": "",
"unverified": true
},
"overall_confidence": { "grade": "", "why": "" },
"headline": "e.g. Automating these unlocks ~12% of annual capacity (~$Xk/yr), low/base/high",
"shown_caveat": ""
},
"processes": [
{
"name": "",
"category": "",
"role_department": "",
"selected_for_detail": true,
"tier": "process",
"tier_note": "process | candidate (low confidence, sub-threshold but recurring); reason if candidate",
"frequency": {
"value": "",
"period": "",
"basis": "reasoned",
"evidence": [],
"annualization_multiplier": "",
"id_corroborated_runs": "K of N"
},
"time_per_instance_min": {
"n": 0,
"median": 0,
"iqr": "",
"basis": "reasoned",
"evidence": []
},
"annual_hours": { "low": 0, "base": 0, "high": 0, "basis": "estimated" },
"common_path": {
"description": "",
"approx_coverage_pct": 0,
"basis": "estimated",
"note": ""
},
"steps": [
{
"action": "",
"app": "",
"input": "",
"output": "",
"time_min": 0,
"pain": "",
"automatable": true,
"note": "",
"basis": "reasoned",
"evidence": []
}
],
"decision_points": [{ "question": "", "branches": [], "basis": "estimated" }],
"time_per_app_per_run": [{ "app": "", "minutes": 0 }],
"reconciliation": {
"per_run_min": 0,
"frequency_per_year": 0,
"automatable_pct": 0,
"loaded_rate": 0,
"hours_freed": 0,
"annual_savings": 0,
"note": "per_run_min x (frequency_per_year/60... ) -> the displayed automatable_pct reproduces annual_savings; downstream reads these, never re-derives"
},
"what_varies": "",
"what_constant": "",
"automation": {
"verdict": "kill|leave_as_is|redesign|automate",
"future_state": "no_change|ai_augmented|human_in_the_loop|automated|eliminated",
"route": "current_stack|internal_app|claude_cowork|current_automation_tool|n8n|gumloop|other",
"effort_tier": "quick_win|extra_engagement",
"work_nature": "mechanical|creative_judgment",
"mechanism": "",
"removes_manual_steps": [],
"feasibility": "api|webhook|agent|none",
"tool": "",
"approach": "",
"automatable_fraction_low": 0.0,
"automatable_fraction_high": 0.0,
"human_remainder": "",
"automation_cost": {
"build_hours": "",
"tool_monthly": 0,
"learning": "low|med|high",
"maintenance": "low|med|high"
},
"basis": "estimated"
},
"scores": { "impact_hours": 0, "automatability": 0, "effort": "", "priority_rank": 0 },
"confidence": { "grade": "", "drivers": "" },
"savings": {
"annual_hours_saved": { "low": 0, "base": 0, "high": 0 },
"gross_annual_dollars_saved": { "low": 0, "base": 0, "high": 0 },
"net_annual_dollars_saved": { "low": 0, "base": 0, "high": 0 },
"payback_months": 0,
"formula": "",
"inputs": [],
"dominant_uncertainty": ""
}
}
],
"systems_landscape": [
{
"system": "",
"function_group": "",
"time_share": "",
"roles": [],
"automation_status": "",
"basis": "reasoned"
}
],
"patterns_cross_check": [
{ "pattern": "", "their_frequency": "", "our_frequency": "", "discrepancy_note": "" }
],
"reasoning_log": [
{
"candidate": "",
"verdict": "process|not_a_process|candidate",
"why": "",
"common_path_coverage_pct": 0,
"case_n": 0,
"evidence": []
}
],
"assumptions": [
{
"plain": "",
"process": "",
"observed_count": "",
"window": "",
"assumed_cadence": "",
"annual_estimate": ""
}
],
"connector_checks": [
{ "server": "", "read": false, "write": false, "verified_via": "", "date": "" }
],
"cowork_skills": [
{
"process": "",
"name": "",
"benefit_line": "",
"trigger": "",
"steps": { "connect": "", "create": "", "run": "" },
"notes": "",
"connectors": [],
"human_approval_gate": ""
}
],
"bottom_line": {
"automatable_task_count": 0,
"hours_saved_per_year": { "low": 0, "base": 0, "high": 0 },
"dollars_saved_per_year_net": { "low": 0, "base": 0, "high": 0 },
"capacity_unlocked_pct_per_person": 0,
"capacity_unlocked_pct_per_role": 0,
"how": ""
},
"what_would_sharpen": [],
"narrative_summary": "",
"review": {
"verdict": "",
"checkpoint": "approved|adjusted|auto-passed",
"demoted": [],
"stripped": [],
"double_count_check": "passed",
"threshold_sensitivity": "",
"reproducibility_note": "deterministic numbers reproduce exactly; AI-judgment calls may vary slightly"
}
}
process-analysis.md mirrors this readably: coverage header (with the n=1 line), baseline, exec roll-up (ranges + plain-word labels + shown_caveat), then the full Checkpoint triage table (every qualifying process AND every candidate-tier row, confirmed processes first sorted by impact then occurrence, candidates below: name, one-sentence what-it-is, annual hours, provisional automatable %, difficulty, uplift, tier + confidence) so nothing found is hidden, a detailed section per selected process (common-path summary, steps table, automation, savings ranges + formula + dominant uncertainty) with tabled-only ones left as triage rows, systems landscape, patterns cross-check, confidence per process + overall, a short "what would sharpen this" list, then reasoning log + assumptions. Then a Bottom line block: count of automatable tasks, time + net $ saved/yr, efficiency unlock per person and per role, and one or two plain sentences on how it's done (mirrors bottom_line). End with a Storytelling summary: under 100 words, plain English, grounded only in the ledger numbers + caveats, covering what we found, what to do, the benefit, and why (mirrors narrative_summary). Exec roll-up aggregated + scoped n=1; processes per role/department; no personal names.
Cowork skills (deliverable): for each automatable process, emit a ready-to-run skill spec in cowork_skills (name, benefit_line, trigger, steps, notes, connectors, a human-approval gate), mirroring the MemoryLane Skill tab. Write the steps in a non-technical 3-step mold a non-technical owner understands: Connect (which apps to connect to Claude), Create (the skill that drafts the output), Run (the cadence, that they click to run it, and that a human reviews and approves before anything is written). Lead each with one plain benefit_line in the form "Use AI to [draft X] for human review instead of [manual grind]". Keep it jargon-free, frame skill setup as done-for-them, and never require the reader to know how to build a skill. Optionally write a standalone SKILL.md for the top one to the working dir.
Map-ready: each process's steps + decision_points + time_per_app_per_run are everything a downstream map skill needs to draw a node-per-step map; this skill emits the data, rendering is downstream and self-contained.
Hand-off contract: every exec-bound number carries its basis label + a one-line shown_caveat; the packaging skill must display both adjacent to each headline and reproduce the reasoning-log verdict count. Each process also carries the reconciliation block (per-run time, frequency/yr, displayed automatable %, loaded rate, hours freed, annual savings) so the report reconciles every figure across pages by reading, never re-deriving; the displayed automatable % is the realized fraction (per Stage 6). (A contract, not enforceable on a sibling renderer; Final Review asserts the data carries it.) Emit one source value per metric at a consistent rounded grain, and ensure per-process figures sum to their role/department subtotal and the exec total, so the report reconciles without hand-edits.
⛔ Stop at the ledger. Do NOT package. Your deliverable ENDS at
process-analysis.{md,json} + the 5-line chat summary (+ optional cowork_skills
specs). Do not render HTML, do not create a PDF, do not invoke or hand straight
into the report-new/packaging skill. Packaging is a separate, user-initiated
step, and that skill shows the HTML deck for review before any PDF exists. After
writing the files, give the short summary and stop; if the user wants the visual
deck, they start packaging next. Never chain analysis → PDF in one run.
get_activity_details (OCR) for high-value confirmations (case IDs, true durations on 2 to 3 sample instances per top process, automation specifics). Keep it bounded.report-new skill; the user starts packaging, and HTML review precedes any PDF (see "Stop at the ledger" in §5).