Rishikesh Ranjan — rishikeshranjan.com

// 9-phase lifecycle

Learn

/ttm-learn

Learn phase: extract lessons from campaign measurement data, propose reference file edits with human approval gates, log root-cause taxonomy entries to LEARNINGS.md, and extract cross-campaign patterns.

phase 9manual invocation
allowed-tools:ReadWriteBashGlobGrepAskUserQuestion
SKILL.md ↗

Overview

ttm-learn runs the Learn phase of the takeToMarket lifecycle. It reads finished campaigns — briefs, gate results, ship records, measurement output — and extracts compound learnings: which positioning angles converted, which channels under- or over-delivered, and which playbook variants worked. Those lessons are logged to .taketomarket/LEARNINGS.md with a structured root-cause taxonomy, and the skill proposes reference file edits behind per-edit human approval gates.

Invoke it with ttm-learn [campaign-slug]. A campaign must already be measured (run ttm-measure first). Future briefs auto-load LEARNINGS.md as Tier 1 context, so each campaign's data biases the next brief toward what actually worked — the marketing equivalent of a postmortem doc plus a regression suite of tactics.

Why it matters: marketing learnings without a structured store dissolve into folklore inside three months. Learn turns each campaign's data into a versioned doc that compounds reference-file improvements and a cross-campaign pattern library.

How it works

The workflow at ${CLAUDE_PLUGIN_ROOT}/workflows/lifecycle/learn.md closes the learning loop: lessons from each campaign compound into reference file improvements and pattern libraries that make future campaigns better. LEARNINGS.md is already loaded as Tier 1 context in the brief workflow (LRNG-04), so future campaigns automatically benefit from past learnings. It proposes reference file edits as narratives with per-edit approval gates (D-07, D-08), logs root-cause taxonomy entries (D-10, LIFE-17), and runs pattern extraction across campaigns once 3+ campaigns have lessons (LRNG-03).

Required reading

  • ${CLAUDE_PLUGIN_ROOT}/references/context-loading.md
  • ${CLAUDE_PLUGIN_ROOT}/references/learnings-extraction.md

Constraints

Reference file edit targets

Only the following files may be edited via this workflow (per D-09):

  • .taketomarket/BRAND.md
  • .taketomarket/ICP.md
  • .taketomarket/CHANNELS.md
  • .taketomarket/POSITIONING.md (via /ttm-positioning-shift only — never directly)
  • .taketomarket/METRICS.md
  • .taketomarket/COMPETITORS.md

Append-only LEARNINGS.md

Lessons are appended after the <!-- LESSONS BELOW THIS LINE --> marker. Existing lesson rows must never be deleted or modified. Summary counters are updated atomically with the append.

Text-mode detection

Text mode (--text flag): Set TEXT_MODE=true if --text is present in $ARGUMENTS or if the runtime is not Claude Code. When TEXT_MODE is active, replace every AskUserQuestion call with a plain-text numbered list.

if echo "$ARGUMENTS" | grep -q -- '--text'; then TEXT_MODE=true; fi

If the AskUserQuestion tool is not available in the current runtime, set TEXT_MODE=true. When TEXT_MODE is active, replace each AskUserQuestion with a plain-text numbered list:

[HEADER]
[QUESTION]
  1. [OPTION_1_LABEL] -- [OPTION_1_DESCRIPTION]
  2. [OPTION_2_LABEL] -- [OPTION_2_DESCRIPTION]
  ...
Type the number of your choice:

Step 0: First-run inline education

Read .taketomarket/CONFIG.md. Parse first_run_seen (object) and inline_education (boolean, default true).

If inline_education is false, skip this step. Else if first_run_seen.ttm-learn is not true, print the explainer below verbatim, then mark this skill as seen:

node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" first-run mark ttm-learn

Use this exact check (bash) to decide whether to print: node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" first-run check ttm-learn --raw — the JSON seen field is true once the explainer has run before.

This skill is backed by the bundled script ${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs, which handles first-run tracking and campaign state reads/writes.

Explainer for /ttm-learn

/ttm-learn reads finished campaigns — briefs, gate results, ship records, measurement output — and extracts compound learnings into .taketomarket/LEARNINGS.md: what positioning angles converted, which channels under- or over-delivered, which playbook variants worked. Future briefs auto-load this file.

Why it matters: marketing learnings without a structured store dissolve into folklore inside three months. Learn turns each campaign's data into a versioned doc that biases the next brief toward what actually worked — the marketing equivalent of a postmortem doc plus a regression suite of tactics.

(Canonical source: references/inline-education-blurbs.md. Embedded verbatim because workflows do not @-resolve files at runtime.)

Step 1: Load context

takeToMarket > LOADING CAMPAIGN CONTEXT FOR LEARNING

Extract SLUG from $ARGUMENTS (strip the --text flag if present):

SLUG=$(echo "$ARGUMENTS" | sed 's/--text//g' | xargs)

If SLUG is empty, error: "Usage: /ttm-learn [campaign-slug]. Provide a campaign slug." Exit.

Load Tier 1 summaries from all 9 reference files (lines 1 to <!-- END_SUMMARY -->):

  • .taketomarket/POSITIONING.md
  • .taketomarket/BRAND.md
  • .taketomarket/ICP.md
  • .taketomarket/CHANNELS.md
  • .taketomarket/STATE.md (frontmatter only)
  • .taketomarket/CALENDAR.md
  • .taketomarket/COMPETITORS.md
  • .taketomarket/METRICS.md
  • .taketomarket/LEARNINGS.md

Load Tier 2 (full content) for learning analysis:

  • .taketomarket/LEARNINGS.md (existing lessons and patterns)
  • .taketomarket/METRICS.md (metric definitions for delta interpretation)
  • .taketomarket/BRAND.md (brand guidelines for edit proposals)
  • .taketomarket/ICP.md (ICP data for edit proposals)
  • .taketomarket/CHANNELS.md (channel data for edit proposals)
  • .taketomarket/COMPETITORS.md (competitor data for edit proposals)
  • .taketomarket/POSITIONING.md (read-only — for positioning drift detection)

Load campaign artifacts per the scan order from learnings-extraction.md:

  • .taketomarket/CAMPAIGNS/${SLUG}/STATE.md (frontmatter for gate results, run counts)
  • .taketomarket/CAMPAIGNS/${SLUG}/MEASUREMENT.md (measurement report with outcome data)
  • .taketomarket/CAMPAIGNS/${SLUG}/BRIEF.md (original strategy and targets)
  • .taketomarket/CAMPAIGNS/${SLUG}/VERIFICATION.md (gate details if it exists)
  • .taketomarket/CAMPAIGNS/${SLUG}/FIX-BRIEF-*.md (fix details if they exist)
  • .taketomarket/CAMPAIGNS/${SLUG}/REVIEW-FEEDBACK-*.md (reviewer comments if they exist)

If .taketomarket/CAMPAIGNS/${SLUG}/ does not exist, error: "Campaign '${SLUG}' not found. Check the slug and try again." Exit.

If .taketomarket/CAMPAIGNS/${SLUG}/MEASUREMENT.md does not exist, error: "No measurement data found for '${SLUG}'. Run /ttm-measure first." Exit.

Step 2: Validate campaign state

node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign state ${SLUG} --raw
  • Verify campaign is in measured or learned phase (re-learning is allowed)
  • If not measured, display an error and exit:
takeToMarket > ERROR
Campaign must be measured before learning. Current phase: ${PHASE}.
Run /ttm-measure first.

Step 3: Extract outcome deltas (per LRNG-02)

Compare measurement results from MEASUREMENT.md against brief targets from BRIEF.md:

  • For each metric in MEASUREMENT.md, find the corresponding target from BRIEF.md
  • Calculate delta: ((actual - target) / target) * 100
  • Classify each delta:
    • overperformed: more than 10% above target
    • met: within +/-10% of target
    • underperformed: more than 10% below target

Display:

takeToMarket > OUTCOME DELTAS

| Metric | Target | Actual | Delta | Classification |
|--------|--------|--------|-------|----------------|
| [outcome metric] | [target] | [actual] | [+/-X%] | [over/met/under] |
| [output metric 1] | ... | ... | ... | ... |

Step 4: Classify lessons per root-cause taxonomy (per LRNG-01, LIFE-17)

For each underperforming or failing element, classify using the 7-category root-cause taxonomy from learnings-extraction.md:

  • positioning-drift — asset deviated from POSITIONING.md (used competitor language, wrong value prop, or off-brand messaging)
  • weak-hook — opening failed to capture attention (generic intro, no specific pain addressed, buried lede)
  • wrong-channel — content published on ineffective channel for target ICP or content type
  • bad-timing — published at wrong time, wrong cadence, or conflicted with external events
  • unverifiable-claim — claim lacked proof point, source, or measurable evidence
  • broken-funnel — CTA or conversion path was broken, misaligned, or missing
  • creative-fatigue — repeated format or angle with diminishing returns (audience saturation)

For each overperforming element, classify as success with a description of what pattern drove the win.

Artifact scan order (per learnings-extraction.md):

  1. STATE.md gate fields for pass/fail patterns
  2. VERIFICATION.md for specific gate failure details
  3. FIX-BRIEF files for root-cause analysis already performed
  4. REVIEW-FEEDBACK files for human reviewer signals
  5. BRIEF.md for strategic context

Generate 1 lesson per notable delta (both successes and failures). Each lesson is one specific, actionable sentence — not a generic observation.

Display each lesson:

takeToMarket > LESSONS EXTRACTED

${N} lessons identified:

1. [${CATEGORY}] ${LESSON_TEXT}
2. [${CATEGORY}] ${LESSON_TEXT}
...

Step 5: Propose reference file edits (per D-07, D-08, D-09, LIFE-16)

For each lesson that implies a systemic change to a reference file, propose an edit using the narrative + apply approach (per D-07).

If no lessons imply systemic changes, display:

takeToMarket > REFERENCE FILE EDITS
No systemic reference file edits identified. All lessons logged as observations.

5a. Present the narrative

Display the reasoning in plain language:

takeToMarket > PROPOSED EDIT ${N} of ${TOTAL}

Based on campaign "${SLUG}" results, [specific observation from measurement data].
[Reasoning derived from the data and campaign artifacts].
Therefore, [specific edit proposal describing the change].

File: .taketomarket/${TARGET_FILE}.md
Section: [section name where the edit applies]
Change: [specific addition, modification, or removal]

5b. Ask for approval (per D-08)

Each edit gets its own approval gate. No batch-apply.

If AskUserQuestion is available (TEXT_MODE=false): Use AskUserQuestion with options:

  • Apply — Update ${TARGET_FILE}.md with this change
  • Skip — Keep ${TARGET_FILE}.md unchanged, log lesson only
  • Modify — Edit the proposed change before applying

If TEXT_MODE=true:

Apply this edit to .taketomarket/${TARGET_FILE}.md?
  1. Apply -- Update ${TARGET_FILE}.md with this change
  2. Skip -- Keep ${TARGET_FILE}.md unchanged, log lesson only
  3. Modify -- Edit the proposed change before applying
Type the number of your choice:

5c. POSITIONING.md special handling (per D-09)

If the proposed edit targets POSITIONING.md, do NOT offer direct apply. Instead:

If AskUserQuestion is available (TEXT_MODE=false): Use AskUserQuestion with header "POSITIONING.md Edit Detected":

  • Launch positioning shift — Start /ttm-positioning-shift with this proposal
  • Skip — Log the lesson but do not change POSITIONING.md

If TEXT_MODE=true:

This edit targets POSITIONING.md, which is locked during campaigns.
This change requires /ttm-positioning-shift.
  1. Launch positioning shift -- Start /ttm-positioning-shift with this proposal
  2. Skip -- Log the lesson but do not change POSITIONING.md
Type the number of your choice:

5d. Apply the user's choice

  • If user selects "Apply": Read the target file, find the specified section, apply the edit, write the updated file. Track as edits_applied.
  • If user selects "Modify": Ask user for the modified version (via AskUserQuestion or text prompt), then apply the modified edit. Track as edits_applied.
  • If user selects "Skip": Set Action Taken for this lesson to "none — skipped by user."
  • If user selects "Launch positioning shift": Note in the lesson log Action Taken: "Deferred to /ttm-positioning-shift." Do NOT apply the edit.

Track counts: edits_proposed (total proposed), edits_applied (Apply + Modify selections).

Step 6: Append lessons to LEARNINGS.md (per LRNG-01, LIFE-17)

  1. Read .taketomarket/LEARNINGS.md
  2. Find the marker line: <!-- LESSONS BELOW THIS LINE -->
  3. Insert new lesson rows immediately after the marker, one row per lesson from Step 4:
| ${DATE} | ${SLUG} | ${CATEGORY} | ${LESSON_TEXT} | ${ACTION_TAKEN} |

Where:

  • ${DATE} is today's ISO date (e.g., 2024-03-15)
  • ${SLUG} is the campaign slug
  • ${CATEGORY} is the root-cause taxonomy category (or success)
  • ${LESSON_TEXT} is the one-sentence lesson
  • ${ACTION_TAKEN} is what was done (e.g., "Updated BRAND.md hook guidelines", "none — skipped by user", "Deferred to /ttm-positioning-shift")

Update the Summary section counters at the top of the file:

  • Total lessons: increment by number of new lessons
  • Last lesson date: set to today's date
  • Top pattern: recalculate the most frequent category across all lesson rows

Write updated LEARNINGS.md.

Step 7: Pattern extraction (per LRNG-03)

Count the number of distinct campaign slugs in the LEARNINGS.md Lessons Log table (the Campaign column).

If fewer than 3 campaigns have entries:

takeToMarket > PATTERN EXTRACTION

Pattern extraction requires 3+ campaigns with lessons. Current: ${N}.
Skipping pattern extraction -- will run automatically after ${3 - N} more campaigns.

If 3 or more campaigns have entries: scan all lesson rows in the Lessons Log for frequency patterns across campaigns:

  1. Winning Hooks: Identify hooks or openings that appeared in success lessons across 2+ campaigns. Extract the common pattern or approach.
  2. Winning Angles: Identify positioning angles or themes that showed consistent overperformance across campaigns.
  3. Winning Formats: Identify content formats that produced the best outcome metric results across campaigns.
  4. Anti-Patterns: Identify approaches that appeared in failure lessons (any root-cause category except success) across 2+ campaigns.

Update the Pattern Extraction sections in LEARNINGS.md:

  • Replace the [Populated after 3+ campaigns...] placeholder text with actual patterns
  • Each pattern entry should include: the pattern description, which campaigns exhibited it, and the measured impact

Write updated LEARNINGS.md.

Step 8: Update campaign state

node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} phase learned
node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} phase.learned "${ISO_DATE}"
node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} learn.run_count ${RUN_COUNT}
node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} learn.last_run ${ISO_DATE}
node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} learn.lessons_extracted ${LESSON_COUNT}
node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} learn.edits_proposed ${PROPOSED_COUNT}
node "${CLAUDE_PLUGIN_ROOT}/bin/ttm-tools.cjs" campaign update ${SLUG} learn.edits_applied ${APPLIED_COUNT}

Where:

  • ${RUN_COUNT} is the previous learn.run_count + 1 (or 1 if first run)
  • ${ISO_DATE} is today's ISO date
  • ${LESSON_COUNT} is the number of lessons extracted in Step 4
  • ${PROPOSED_COUNT} is the number of reference file edits proposed in Step 5
  • ${APPLIED_COUNT} is the number of edits the user approved (Apply + Modify)

Step 9: Display summary and next steps

takeToMarket > LEARN PHASE COMPLETE

Campaign: ${SLUG}
Lessons extracted: ${LESSON_COUNT} (${SUCCESS_COUNT} successes, ${FAILURE_COUNT} failures)
Reference edits proposed: ${PROPOSED_COUNT}
Reference edits applied: ${APPLIED_COUNT}
Pattern extraction: ${PATTERN_STATUS}

Lessons appended to: .taketomarket/LEARNINGS.md

Next: Run /ttm-archive ${SLUG} to finalize the campaign, or start a new campaign.
Future campaigns will load these lessons via /ttm-brief to prevent repeating mistakes.

Where ${PATTERN_STATUS} is one of:

  • "Completed — ${PATTERN_COUNT} patterns identified" (if 3+ campaigns)
  • "Skipped — ${N} of 3 campaigns needed" (if fewer than 3)

Checklist

  • Campaign is in measured or learned phase
  • Campaign artifacts scanned in correct order (STATE > VERIFICATION > FIX-BRIEF > REVIEW-FEEDBACK > BRIEF)
  • Outcome deltas calculated for all metrics
  • Each lesson classified with root-cause taxonomy category
  • Reference file edits proposed as narratives with per-edit human approval
  • POSITIONING.md edits routed through /ttm-positioning-shift (not direct edit)
  • Lessons appended to LEARNINGS.md after marker line
  • Summary counters updated in LEARNINGS.md
  • Pattern extraction runs only if 3+ campaign slugs in lessons log
  • Campaign state updated with learn.* fields