Pmm okrs

pmm-okrs

A guided OKR builder for Product Marketing teams. Run it at the start of every quarter. Outputs a complete, review-ready OKR set you can paste directly into the PMM OKR Builder sheet. Learns from prior quarters to improve confidence calibration and KR quality.

Trigger

• When: Start of any quarter when setting PMM OKRs. When reviewing or stress-testing existing KRs before committing. When building a measurement plan or leadership-ready exec narrative from a finalised OKR set.
• Not for: Company-level OKR design (not PMM-specific) → general planning tool. Revenue forecasting or headcount planning. OKR tooling setup (Lattice, Workday) — this skill produces content, not configuration. If no quarterly strategy exists yet → run hs-pmm-strategy first, then return here.
• Example prompts:
  • "Help me set our Q3 OKRs"
  • "Are these KRs measurable enough?"
  • "Build a scorecard for my chosen option"
  • "Write OKRs for my team lead who owns competitive intelligence"
  • "Stress-test this KR: improve win rate in enterprise"
  • "I need to present our goals to the exec team next week"

---

Inputs

• Args: Company objective, PMM mandate, team size, primary metric, biggest challenge, ICP, and named competitors. All optional at start — skill gathers via intake flow.
• Defaults: If no args provided, run intake flow via /build. If partial context is provided, infer where possible and surface gaps explicitly before proceeding.
• Context keys:
  • .agents/product-marketing-context.md — optional but recommended. Load Revenue Levers, Goals & KPIs, Big Bet Campaigns, Company Overview silently if present.
  • /context/meta-patterns.md — optional; recurring patterns from all skills (guardrail prompts)
  • knowledge/okrs/rules.md — apply confirmed OKR craft rules by default.
  • knowledge/okrs/hypotheses.md — test any active hypothesis if applicable today.
  • decisions/ — check for prior decisions before making new recommendations.
  • /context/skill-sessions.md — optional; prior quarter OKR data for confidence calibration Brain contract: Reads: Section 2 (ICP), Section 3 (Positioning), Section 5 (Revenue), Section 6 (Goals). Writes: /context/skill-sessions.md, /foundation/brain.md Section 5 (if lever weights change).

---

Pre-flight

Load guardrails first: Check /context/meta-patterns.md for recurring OKR patterns. If pattern matches (e.g., "confidence calibration off by 15% in Q2-Q3"), surface guardrail prompt before Step 1.

Before starting, check .agents/product-marketing-context.md. If it exists — load silently:

• ## Revenue Levers → align OKRs to the stack-ranked levers
• ## Goals & KPIs → use North Star + OMTM as anchors
• ## Big Bet Campaigns → surface as project goals
• ## Company Overview → stage and business model context Confidence awareness: If loaded sections are 🔴, flag before building OKRs:

"Revenue Levers is marked as Placeholder — OKRs built on this may need revisiting. Want to update it first?" If missing: Proceed. Surface once: "Run hs-product-marketing-context BUILD first for sharper OKRs. Continuing."

Load prior quarter data: Check /context/skill-sessions.md for last quarter's OKR results to calibrate confidence:

"Last quarter: confidence was 75%, actual achievement was 78%. You're well-calibrated. Recommend similar range this quarter."

Related skills — cross-reference before or after this skill:

• hs-pmm-strategy → run before this skill if no quarterly strategy exists yet
• hs-product-requirement-doc → PRDs inform project OKRs; check for alignment
• hs-gaccs-brief → campaign briefs should trace back to OKR project goals
• meta-synthesis → after 3+ quarters logged, meta-synthesis detects OKR patterns

---

Steps

Step 0: Surface Guardrails (NEW)

Before intake, check for patterns:

If /context/meta-patterns.md exists and contains OKR patterns:

🔁 PATTERN DETECTED FROM PRIOR QUARTERS

I've detected [specific pattern] in [N] prior quarters.
Example: "Confidence off by 15%", "External dependency blockers 3+ times", "KR design failure on 'improve adoption'"

Quick question: Are you seeing this pattern again in Q4?
- If YES → we'll flag it as a confirmed problem + propose a fix
- If NO → we'll watch for other patterns

This helps calibrate your confidence and KR design.

Common guardrail patterns to surface:

• "Confidence too high (90%+) but achievement 65%" → "Recommend: set confidence 70% or below"
• "External dependencies blocked 3+ KRs" → "Build dependency risk into confidence"
• "KR too vague again ('improve adoption')" → "Be specific: 'Adoption in SMB segment 60% → 75%'"
• "Lever weight assumptions changed mid-quarter" → "Lock lever weights at start"

If patterns apply, ask guardrail question. User can skip, but they've been warned.

Step 1 — /build

1. Run intake (or infer from pasted context).
2. Load knowledge/okrs/rules.md + check guardrails from /context/meta-patterns.md.
3. Check decisions/ for prior choices in this area.
4. Check /context/skill-sessions.md for prior quarter confidence calibration.
5. Generate three OKR options.
6. Run independent evaluation pass (Block 3) on all three.
7. Present with Quality Gate results inline + confidence recommendations based on prior quarters.
8. Log option selection to decisions/.

Step 2 — /review

1. Accept pasted OKRs.
2. Run each KR through all five Quality Gates (binary).
3. Flag every failure with an ADVERSARIAL CALLOUT and a rewrite.
4. Return annotated set.

Step 3 — /scorecard

1. Work from OKRs in session or ask user to paste.
2. Map each KR to metric, target, measurement method.
3. Group by category.
4. Confirm Weight = 100%.
5. Output scorecard.

Step 4 — /exec

1. Confirm OKRs are finalised (not drafts).
2. Translate to one-paragraph exec narrative.

Step 5 — /map

1. For each KR, generate required projects with owner type, effort (S/M/L), timeline.
2. Flag capacity conflicts if team size known.
3. Cross-reference against guardrails: "External dependencies flagged in 3 prior quarters. This KR has 2. Accept risk?"

Step 6 — /stress-test [KR]

1. Accept one KR.
2. Run through five Quality Gates.
3. Return per-gate pass/fail + rewrite.

Step 7: Post-Session Logging (NEW)

After every session, log structured data to /context/skill-sessions.md:

skill: pmm-okrs
session_date: 2026-06-21
quarter: "Q3 2026"
okr_set_version: 1
objectives_count: 3
key_results_count: 9
confidence_level: "medium"
confidence_range: "65-70%"
krs_with_baseline_metrics: 9
krs_with_tracking_mechanism: 7
krs_with_stretch_factor: 6
aggressive_vs_conservative: "mixed"
dependencies_identified: true
dependencies_external: 3
dependencies_internal: 1
prior_quarter_okrs:
  - quarter: "Q2 2026"
    krs_achieved: 7
    krs_partial: 1
    krs_missed: 1
    achievement_rate: 0.78
    confidence_predicted: 0.75
current_vs_prior_stretch: "more_aggressive"
confidence_vs_last_quarter: "higher"
confidence_calibration_delta: "+2%"
output_path: "/foundation/okrs/Q3-2026-PMM-OKRs.md"
guardrails_triggered:
  - "External dependencies: 3 (same as Q2). Last quarter we missed 1 KR due to external dependency. Recommend: align with dependent teams by Week 1 of Q3."
  - "Stretch factor: 6/9 KRs are ambitious (2x+). This is higher than Q2 (4/8). Recommend: ensure team capacity supports this."
  - "Confidence calibration: Q2 predicted 75%, achieved 78%. You're well-calibrated. Recommend similar range (72-78%) for Q3."
brain_updates_proposed: []

This feeds into meta-synthesis skill (monthly) which detects OKR patterns across quarters and updates guardrails.

Step 8: Deliver Output and Log Learning

Deliver the OKR set. Then run the self-improvement loop: write session file → update knowledge base → log decisions → run quality gate → propose confidence adjustments.

Outputs

• Files written:

  • /context/skill-sessions.md — row appended with session metadata and guardrails (NEW)
  • decisions/YYYY-MM-DD-{topic}.md when a strategic OKR choice is logged.
  • knowledge/okrs/hypotheses.md and rules.md when the self-improvement loop triggers at session end.

• Chat output format: Three OKR option blocks in code-fence structured output, each with Quality Gate results inline. Scorecard table. Exec Summary paragraph. All formatted for direct paste into the PMM OKR Builder spreadsheet.

• External side effects: None beyond context writes.

---

Verification

• Guardrails checked before intake (Step 0) — patterns from prior quarters surfaced.
• All /build output contains three OKR options unless user explicitly requests fewer.
• Every option includes Quality Gate results (five binary checks) before delivery.
• No output delivered before the independent evaluation pass (Block 3) has run.
• Adversarial callouts surface inline before delivery, never post-delivery.
• Decision log written whenever a recommendation will affect the user's quarter.
• /exec output produced only from finalised OKRs, not from draft options.
• Scorecard Weight confirmed at 100% before delivery.
• Confidence calibration checked against /context/skill-sessions.md prior quarter data.
• Session logged to /context/skill-sessions.md with all metadata.

---

Do Not Use For

• hs-pmm-strategy — if no quarterly strategy exists yet, run that first. This skill builds OKRs from a strategy, not instead of one.
• hs-prioritization-frameworks — for prioritising which initiatives to include in a quarter before OKRs are set. Run that upstream, then return here.
• hs-gaccs-brief — for campaign planning that traces back to OKRs already set. Use after /build to brief individual campaigns.
• Company-level OKR design — this skill is PMM-specific. Exec team or company OKRs require different framing and are out of scope.
• OKR tooling setup — this skill produces OKR content, not Lattice/Workday configuration or workflow automation.

---

Reasoning Architecture

Block 1 — Knowledge Architecture (Learning Loop)

Before any task:

1. Load knowledge/INDEX.md and relevant domain folders.
2. Apply rules.md by default.
3. Test any active hypothesis if applicable today.
4. Load /context/meta-patterns.md for cross-skill guardrails.
5. Load /context/skill-sessions.md for prior quarter confidence calibration.

After any session:

1. Extract 1–3 insights.
2. Unconfirmed → hypotheses.md.
3. Confirmed 3+ times → auto-promote to rules.md.
4. Contradicted → demote to hypotheses.md.
5. Log confidence calibration delta for future quarters.

Environment note: Persistent /knowledge/ works in Claude Code and Cowork. In Claude.ai chat, surface insights at end of session for manual carry-forward.

Block 2 — Decision Journal

Check decisions/ before any recommendation. Log new decisions immediately.

File: decisions/YYYY-MM-DD-{topic}.md
Decision / Context / Alternatives / Reasoning / Trade-offs / Supersedes

Log when:

• Choosing between OKR options.
• Recommending a measurement method.
• Flagging a failing gate.
• Any recommendation affecting the user's quarter.
• Confidence calibration adjustments based on prior quarter data.

Block 3 — Independent Evaluation Pass

After generating any output: re-read cold as evaluator. Run all five Quality Gates binary. Rewrite failures before delivery. Report gate results inline.

Confidence calibration check: Compare predicted confidence vs. actual achievement from prior quarter. Adjust recommendations:

Last quarter: Predicted 75%, Achieved 78% → You're well-calibrated
This quarter: Recommend 72-78% confidence range

---

Commands

/build

Builds three OKR options from scratch with Quality Gate results and confidence calibration. Example prompts:

• Help me set our Q3 OKRs. 3-person PMM team, B2B SaaS, company OKR: grow ARR 40%.
• Build OKRs for a solo PMM at a Series B. Challenge: positioning isn't landing.

/review

Audits existing OKRs against all five Quality Gates. Returns exact fixes. Example input:

Objective: Improve go-to-market in mid-market.
KR 1: Launch 4 battlecards. KR 2: Run monthly training. KR 3: Increase pipeline.

/scorecard

Maps each KR to metrics, targets, and measurement methods.

/exec

Generates one-paragraph exec-ready OKR narrative for QBRs and VP presentations.

/map

Builds OKR → Projects table with owner type, effort (S/M/L), and timeline. Cross-references guardrails.

/individual [specialty]

Generates OKRs for an individual PMM contributor.

• /individual positioning · /individual competitive · /individual gtm

/stress-test [KR]

Runs one KR through all five Quality Gates. Returns pass/fail + rewrite.

Output Format

═══════════════════════════════════════════
OPTION [A / B / C] — [Strategic Focus]
═══════════════════════════════════════════
OBJECTIVE: [1–2 sentence qualitative goal.]
KR 1 — [Name]: [Outcome. Target. Deadline. Measurement.]
KR 2 — [Name]: [Outcome. Target. Deadline. Measurement.]
KR 3 — [Name]: [Outcome. Target. Deadline. Measurement.]
CONFIDENCE: [X%]
CONFIDENCE REASONING: [Based on prior quarter calibration: if last quarter predicted 75% achieved 78%, recommend similar range]
CHOOSE THIS WHEN: [1-sentence fit description.]
KEY PROJECTS: 1. [name — KR] 2. [name — KR] 3. [name — KR]
QUALITY GATE RESULTS:
Gate 1 — Outcome not output:           ✅ / ❌
Gate 2 — Measurable without ambiguity: ✅ / ❌
Gate 3 — Causally linked to objective: ✅ / ❌
Gate 4 — 60–70% confidence:            ✅ / ❌
Gate 5 — Three or fewer KRs:           ✅ / ❌
GUARDRAILS: [Cross-skill patterns from prior quarters, if any]
═══════════════════════════════════════════

---

Quality Gates

Gate	Test	Fail	Pass
1	Outcome, not output	"Launch 4 battlecards"	"Win rate up 8%"
2	Measurable without ambiguity	"Improve messaging"	"80% resonance from 50 reviews"
3	Causally linked to objective	PMM doesn't own lever	PMM controls what moves this
4	60–70% confidence	>90% or <50%	Ambitious but achievable
5	Three or fewer KRs	Four or more	Three or fewer
Adversarial callout format:

⚠️ ADVERSARIAL CALLOUT: [Issue] — [Why it's a problem and what to write instead.]

---

Operating Rules

• Guardrails first. Load /context/meta-patterns.md at pre-flight. Surface guardrail prompt if pattern matches this quarter's planning.
• Load brain context before intake. Pre-flight runs silently — never ask for context already loaded.
• Load prior quarter data for confidence calibration. Check /context/skill-sessions.md before confidence recommendations.
• Three options minimum on /build. Choice architecture is the value.
• Independent evaluation pass is non-negotiable. Unreviewed output is not delivered.
• Adversarial callouts surface before delivery, not after. Rewrites happen during generation.
• Decision logging is not optional. Every recommendation affecting the quarter gets logged.
• Writes only to decisions/, knowledge/, and /context/skill-sessions.md. No writes to brain file.
• Confidence range is enforced. >90% or <50% triggers an adversarial callout. Calibrate against prior quarter achievement.
• Gate results in table format only. Binary ✅ / ❌ — no narrative substitution.
• Scorecard Weight confirmed at 100% before delivery. Surface discrepancy if unbalanced.
• /exec only from finalised OKRs. Prompt for option choice if drafts only.
• Always log. Every quarter's OKR session logged to /context/skill-sessions.md. Meta-synthesis learns from achievement rates across quarters.

---

Quality Gate

Runs before final delivery. Score each criterion 1–3. Minimum 17/21 to pass.

Criterion	Standard	Score (1–3)
Guardrails surfaced	/context/meta-patterns.md checked at pre-flight
Confidence calibration	Prior quarter data checked for adjustment recommendations
Three options minimum	All /build delivers 3 options with Quality Gate results
Independent evaluation	All output reviewed cold before delivery
Adversarial callouts	All gate failures surfaced inline with rewrites
Decision logging	Strategic OKR choices logged to decisions/
Scorecard validation	Weight totals 100% confirmed before delivery

On failure: Identify which criterion failed, revise, do not present as final.

Self-Improvement Loop

Before every session:

1. Load guardrails: Check /context/meta-patterns.md for patterns matching this quarter. Surface if found.
2. Load knowledge/INDEX.md and relevant domain folders.
3. Apply rules.md by default.
4. Test any active hypothesis if applicable today.
5. Check decisions/ for prior choices in this area.
6. Load prior quarter data: Check /context/skill-sessions.md for last quarter's OKR results + confidence calibration.
7. Propose confidence range based on prior quarter accuracy.

After every session:

1. Log to /context/skill-sessions.md: Complete session metadata (see Step 7 schema).
2. Extract 1–3 insights.
3. Unconfirmed → hypotheses.md.
4. Confirmed 3+ times → auto-promote to rules.md.
5. Contradicted → demote to hypotheses.md.
6. Log confidence calibration delta (predicted vs. actual achievement).
7. Update knowledge/INDEX.md.

Self-Improvement Trigger format — surface before encoding, never silently:

🔁 SELF-IMPROVEMENT TRIGGER
Pattern: [observed across quarters]
Proposed update: [exact wording]
Location: [file path]
Awaiting approval before encoding.

Confidence Calibration Trigger:

📊 CONFIDENCE CALIBRATION UPDATE
Q2 predicted: 75% | Achieved: 78% | Delta: +3%
Q3 recommendation: Aim for 72-78% confidence range
Reasoning: You're well-calibrated; apply same range this quarter.

---

Changelog

v2.2.0 — 2026-06-21

Added post-session logging + guardrail intake + prior quarter confidence calibration. Guardrails loaded from /context/meta-patterns.md at pre-flight. Step 0 surfaces patterns (e.g., "confidence too high last quarter"). Step 1 now loads /context/skill-sessions.md for prior quarter data. Step 7 logs session metadata to /context/skill-sessions.md for meta-synthesis. Step 5 cross-references guardrails for dependency risks. Output template now includes "CONFIDENCE REASONING" section. Operating Rules reprioritized (guardrails + confidence calibration first). Quality gate expanded to 8 checks. Self-improvement loop now loads guardrails + prior quarter data first.

Changes from v2.1.0:

• Added Step 0: Surface guardrails before intake (reads /context/meta-patterns.md)
• Updated Pre-flight: Load guardrails first + check /context/skill-sessions.md for prior quarter calibration
• Updated Step 1: Load prior quarter data for confidence recommendations
• Updated Step 5: Cross-reference guardrails for dependency risk context
• Added Step 7: Post-session logging to /context/skill-sessions.md
• Updated Output Template: Added "CONFIDENCE REASONING" + "GUARDRAILS" sections
• Updated Outputs: Now writes to /context/skill-sessions.md (NEW)
• Updated Inputs: Now reads /context/meta-patterns.md (NEW) + /context/skill-sessions.md (NEW)
• Updated Operating Rules: "Guardrails first" + "Confidence calibration checked" + "Always log"
• Updated Quality Gate: Added checks for guardrails + confidence calibration (now 8 checks)
• Updated Verification: Added checks for guardrails + confidence calibration + session logging
• Updated Self-Improvement Loop: Load guardrails + prior quarter data first, log confidence delta
• Updated Changelog: Now includes v2.2.0 entry
• Updated metadata: Added logging_enabled: true
• Version bump 2.1.0 → 2.2.0

v2.1.0 — 2026-06-06

Spec compliance pass against SKILL-SPEC v2.0.0. Score: 8/19 → 19/19. Added: Trigger, Inputs, Pre-flight, Steps, Outputs, Verification, Do Not Use For, Operating Rules. Fixed: name field, metadata block, output templates fenced, quality gates as table, self-improvement loop restructured. Trimmed to <500 lines.

v2.0.0 — 2026-05-01

Full rebuild: compounding knowledge graph (Block 1), decision journal (Block 2), independent evaluation pass (Block 3), adversarial callouts, seven commands.

Evals Updates (NEW)

Create: /pmm-execution/skills/pmm-okrs/evals/pmm-okrs.eval.md

# pmm-okrs Evals

## Eval 1: Guardrails surface before intake
**Scenario:** User starts `/build` for Q3. `/context/meta-patterns.md` contains "confidence too high 2x, external dependencies blocked 1x"
**Expected:** Guardrail prompt surfaces before Step 1 intake
**Input:** `/build Q3 2026 for 3-person PMM team`
**Output check:** 
- Does guardrail appear? YES/NO
- Does guardrail match pattern from meta-patterns? YES/NO
- Can user skip guardrail? YES/NO

## Eval 2: Confidence calibration recommended
**Scenario:** `/context/skill-sessions.md` shows Q2: predicted 75%, achieved 78%
**Expected:** Step 1 surfaces recommendation for Q3 confidence range (72-78%)
**Input:** `/build Q3 2026`
**Output check:**
- Confidence reasoning section present? YES/NO
- References Q2 calibration? YES/NO
- Recommends 72-78% range? YES/NO

## Eval 3: Three options delivered with Quality Gates
**Scenario:** User runs `/build` without prior OKRs
**Expected:** Three OKR options, each with 5-gate results (✅/❌ table)
**Input:** `/build Q3 OKRs for SMB focus`
**Output check:**
- Three options present? YES/NO
- Quality Gate table in each option? YES/NO
- Gate results binary (✅/❌)? YES/NO
- No narrative substitutions? YES/NO

## Eval 4: Session logged to /context/skill-sessions.md
**Scenario:** User completes `/build` and `/scorecard`
**Expected:** Entry written to `/context/skill-sessions.md` with full metadata
**Input:** `/build` + `/scorecard`
**Output check:**
- Session logged? YES/NO
- Contains: quarter, objectives_count, krs_with_baseline_metrics? YES/NO
- Contains: prior_quarter_okrs data? YES/NO
- Contains: guardrails_triggered? YES/NO
- Contains: confidence_calibration_delta? YES/NO

## Eval 5: Adversarial callouts surface inline
**Scenario:** User submits OKRs with vague KR ("improve adoption")
**Expected:** Callout surfaces during `/review` with rewrite before delivery
**Input:** `/review` with vague KR
**Output check:**
- Callout appears? YES/NO
- Format: ⚠️ ADVERSARIAL CALLOUT: [Issue]? YES/NO
- Rewrite provided? YES/NO
- Rewrite is specific? YES/NO

## Eval 6: Scorecard Weight confirmed at 100%
**Scenario:** User runs `/scorecard` with weights that don't total 100%
**Expected:** Discrepancy surfaced, scorecard not delivered
**Input:** `/scorecard` with 4 metrics weighted 30%, 30%, 20%, 15%
**Output check:**
- Discrepancy detected? YES/NO
- Totals 95% mentioned? YES/NO
- Scorecard held from delivery? YES/NO
- Prompt to rebalance given? YES/NO

## Eval 7: Prior quarter dependency risks flagged in `/map`
**Scenario:** `/context/skill-sessions.md` shows "external dependencies blocked in Q2"
**Expected:** `/map` output flags this risk for current KRs
**Input:** `/map` with 3 KRs containing external dependencies
**Output check:**
- Guardrail about prior quarter blockers? YES/NO
- Specific to external dependencies? YES/NO
- Asks user to accept risk? YES/NO

## Eval 8: Decision logging on strategic choice
**Scenario:** User selects Option B from `/build` (not Option A)
**Expected:** Decision logged to `decisions/YYYY-MM-DD-{topic}.md`
**Input:** User selects Option B
**Output check:**
- Decision file created? YES/NO
- Contains: Decision / Context / Alternatives / Reasoning? YES/NO
- References Option A as alternative? YES/NO

---

Related Skills

Cross-reference when findings trigger downstream work:

• hs-pmm-strategy → run before this skill if no quarterly strategy exists
• hs-product-requirement-doc → PRDs inform project OKRs; check alignment
• hs-gaccs-brief → campaign briefs should trace back to OKR projects
• meta-synthesis → after 3+ quarters logged, detects OKR patterns + confidence trends
• hs-retro → after quarter closes, retro compares actual vs. predicted OKRs