agentleFS
iwesysiwesys/IWE/.claude/skills/bottleneck-pick/SKILL.md

bottleneck-pick

Skill0 starsChanged 25 days agoMITView on GitHub

Constraint Analyst (DP.ROLE.054): identifies the primary constraint

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Test: if the Constraint Brief contains no references to specific DP.SC items that "are not working" — Phase 2 has failed.

    line 46
  2. Distinctive:

    Strict template principle: all 4 categories are always shown, even when empty ("—").

    line 275
  3. Distinctive:

    After 5–10 runs → analyze precision on Week Close, recalibrate signal weights.

    line 452

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Skill: /bottleneck-pick

see DP.SC.045, DP.ROLE.054, DP.WP.016

TOC methodology carrier for the pilot: Goldratt Five Focusing Steps + Tendon TameFlow Replenishment Cycle + Dettmer Thinking Processes. Implements the role of Constraint Analyst (DP.ROLE.054) through a five-phase VDV cascade. Source: WP-313 (Phase 1 research: Tendon "Tame Your Work Flow", Dettmer, Schragenheim, Goldratt S&T; Phase 11 IntegrationGate).

Triggers

  • User opens an umbrella WP, epic, or project and asks "what to do first"
  • Request: "pick bottleneck", "what is more important", "where to start"
  • Strategic session selecting the critical constraint ("which direction is the bottleneck right now")
  • Diagnostician (R28) requests analysis for a pilot learning pipeline
  • Explicit call /bottleneck-pick --target WP-NNN

SC-first principle (primary rule)

Document-centric approach is an anti-pattern. Starting from a list of pending WPs risks mistaking the delivery channel for the product.

Correct: Phase 2 scans functional promises (DP.SC) BEFORE Phase 3 (signal-scan over the work structure). Test: if the Constraint Brief contains no references to specific DP.SC items that "are not working" — Phase 2 has failed.

API

/bottleneck-pick --target <ref> [--layer intra|platform] [--horizon <horizon>] [--depth <1|2|3>] [--scope <direct|direct+related|full>]

Parameters

ParameterDefaultDescription
--targetrequiredWP-NNN (umbrella WP), weekplan (current week), pilot:<account_id> (pilot learning pipeline), b2:aisystant (full ecosystem — requires --layer=platform), c2:platform (technical layer only — requires --layer=platform), project-name, repo-path
--layerintraintra = within a single system (constraint = phase / WP / block). platform = across C2 subsystems (constraint = subsystem / handover / B3 role). Requires a user-provided systems map.
--systems-mapmemory/project_iwe_systems_map.md in the installed workspacePath to the user-provided map for --layer=platform; an absolute path to a private directory is accepted. The map is not shipped with the public template.
--horizonallday, week, month, wave-1, wave-2, quarter, next-stage
--depth21 = Five Steps only; 2 = + EC for conflicting candidates; 3 = + Coupling-analysis SOTA.011 for top-3 handovers (--layer=platform only)
--scopedirect+relateddirect = target only; direct+related = + related WPs from frontmatter; full = + reachable via dependency graph ≤2 hops

Sources by --target type

TargetWhat the skill readsWhere the source lives
WP-NNN${IWE_GOVERNANCE_REPO:-DS-strategy}/inbox/WP-NNN-*.md + related WPs + git last 7dgovernance
weekplan${IWE_GOVERNANCE_REPO:-DS-strategy}/current/WeekPlan W{N}.md + active-wp.md + DP.SC over active WPsgovernance
pilot:<id>learning.activity_log + learning.cp_assessments + learning.bh_metrics (Neon)platform (learning pipeline)
b2:aisystant03-our-systems-map.md (S2R) + DP.MAP.002 + user map from --systems-map + cross-system SCgovernance + Pack + user file
c2:platformDP.MAP.002 (12 subsystems) + new Q2 entries from user mapPack + user file

Error cases

SituationBehavior
Target not found"Could not find <target>. Check the number or path." → STOP
Target is a closed WP"<target> is already closed. Provide an active WP." → STOP
Target is empty (no structure)"No structure to analyze in <target>." → STOP
--layer=platform, map is missing, empty, or unreadableRun "$IWE_TEMPLATE/scripts/check-platform-systems-map.sh" from the installed template, display its exact diagnostic output, and STOP before the System Card and conclusions. The user must create their own map or provide --systems-map <path>.
Data is stale (>7 days without git activity)Add ⚠️ to signal-scan; do not stop
EC does not converge (empty assumptions)Fall back to Five Steps; note in output
All candidates are not agent-actionable"All candidates are not agent-actionable. A different entry point is needed."

Algorithm — five-phase VDV cascade

VDV principle (DP.M.060): the output of each phase is the input to the next. Every input and output is a physical Artifact.

Phase 1 — Identify system

Input: target-ref + --layer from the user.

Action: 0. For --layer=platform, before analysis, determine the installed workspace root (WS="${IWE_WORKSPACE:-$PWD}"), load . "$WS/.iwe-paths", and verify that IWE_TEMPLATE/IWE_WORKSPACE are non-empty; if the file or variables are missing — report a path setup error and STOP. Then run bash "$IWE_TEMPLATE/scripts/check-platform-systems-map.sh" or, if the parameter is set, bash "$IWE_TEMPLATE/scripts/check-platform-systems-map.sh" --map "<path from --systems-map>". The script lives in the nested template; the map lives in the workspace. Exit code 2: print the diagnostic and STOP — a missing map cannot be replaced by a guess or by DP.MAP.002 alone. The script prints the verified path — use that user file from this point on. For --layer=intra this check is not needed.

  1. Classify the pipeline system type by --target × --layer:
    • Pilot learning pipeline — --target pilot:<id> (FORM.089 RCS, source: Neon learning.*)
    • Work pipeline (intra) — --target WP-NNN | weekplan | project | repo-path, --layer=intra. Structure = directions A–I / phases Phase 1–PhaseN / blocks within a single WP
    • Platform (cross-system) — --target b2:aisystant | c2:platform | WP-NNN-umbrella, --layer=platform. Structure = C2 subsystems + handovers + B3 roles. ⚠️ Validation: the verified user map and DP.MAP.002 must be loaded; take the subsystem count from the map, not from the template.
    • Cohort pipeline — --target cohort:<id> | wave-N
  2. Collect context by type:
    # intra
    git log --oneline --since="7 days ago" <repo>
    cat <target-context-file>
    ls <related-PROCESSES.md>
    
    # platform
    ls PACK-digital-platform/.../08-service-clauses/   # cross-system SC
    cat "<verified_map_path>"                           # path printed by check-platform-systems-map.sh
    git log --oneline --since="7 days ago" --all       # activity across subsystems
  3. Record data freshness. If ≥7 days without activity — ⚠️. For --layer=platform: check DP.MAP.002 freshness (stale >30 days → ⚠️).

Output: System Card (in chat + calibration YAML):

System type: learning_pipeline | work_pipeline_intra | work_pipeline_platform | cohort_pipeline
Target: <target-ref>
Layer: intra | platform
Structure: <directions A–I | phases Phase1–PhaseN | C2 subsystems + handovers | RCS slots>
Subsystems in scope: <N — only when layer=platform>
Data freshness: <date + flags>
Unregistered subsystems: <list — only when layer=platform, if any>

Phase 2 — Scan promises (SC-first)

Input: System Card.

Action:

  1. Locate DP.SC by system type:
    • --layer=intra (WP / epic / repo): SC from WP frontmatter + SC consumed by target roles + SC within a single subsystem
    • --layer=platform (b2 / c2 / umbrella WP): cross-system SC — DP.SC.020 retention, DP.SC.012 onboarding, DP.SC.003 learning, hypothesis G-K1. Additionally check B3-to-B2 SC (team↔ecosystem) — often not formalized; flag if so.
    • Learning pipeline (pilot:): FORM.089 sections (specialization SC)
    • Cohort pipeline: SC from MAP.002 for wave-rollout, adoption, lifecycle
  2. Apply a health test to each SC:
    - "Working?" (event freshness, consumers present, no active incidents)
    - "Broken?" (explicit open incident, no events for N days, negative feedback)
    - "Partial?" (some consumers receive it, others do not)
    For --layer=platform — additional test: "Broken at the handover?" — one subsystem produces output, the other does not accept it (gap between Bot and GitHub App, between Profiler and Stage Evaluator, etc.).
  3. Mark SC-failing and SC-partial items as bottleneck candidates.

Output: SC-status map (format depends on --layer):

For --layer=intra:

| SC ID         | Promise                                  | Status       | Signal                                      |
|---------------|------------------------------------------|--------------|---------------------------------------------|
| DP.SC.020     | Personal guide delivery                  | ⚠️ Partial  | 5/9 pilots have not activated repo          |
| DP.SC.132     | Learner diagnostics                      | ✅ Works    | cp_assessments recorded daily               |
| DP.SC.044     | Event ingest                             | ⚠️ Partial  | subscription.contract_event = 0 rows        |

For --layer=platform (cross-system path):

| SC          | Cross-system path                                  | Status       | Weak link                                       |
|-------------|-----------------------------------------------------|--------------|--------------------------------------------------|
| DP.SC.020   | Bot → /personal-guide-start → GitHub App → repo    | ⚠️ Partial  | handover Bot↔GitHub App (6/9 not activated)     |
| G-K1        | Bot → Onboarding → Profiler → Stage Eval → Nudge   | ❌ FAIL     | Nudge missing (WP-117 not started)               |
| DP.SC.112   | YooKassa → Bot → CRM → Access Manager              | ✅ Works    | stable                                           |

Phase 3 — Identify constraint

Input: SC-status map + work structure from System Card.

Action:

  1. For each SC-failing/partial item and each structure block (direction / phase / WP / subsystem / handover) — signal-scan using TameFlow signals (ordered by predictive value):
#SignalTriggerApplies to
1Queue accumulation rateΔ queue length / 7d ≥30% growthall
2Flow Efficiency droptouch/lead <15% or Δ −10pp/weekall
3Dependency fan-in / handover count>3 handovers in one user journey = policy issueall
4Buffer consumption rateCCPM fever chart: Yellow → monitor, Red → actall
5Policy freeze markersNLP search: "requires approval", "we always do it this way", "the rule is"all
6Cognitive freezegrowing open questions without resolution, signs of procrastinationall
7Cross-system handover counthandovers >3 per user journey across C2 subsystems = systemic--layer=platform only
8Coupling tightness (SOTA.011)SC of one layer depends on the internal implementation of another = tight coupling, candidate handover-constraint--layer=platform only
  1. Select top-3 candidates by combined signal score.
  2. Classify each candidate:
    • Trichotomy (Tendon): Work Flow / Work Process / Work Execution
    • Class: Policy / Resource / Cognitive
    • For --layer=platform — additionally Locus (where the constraint sits): Subsystem (one C2 subsystem is the bottleneck) / Handover (both subsystems work, but SC breaks between them) / Role(B3) (subsystem exists, but the team role is unassigned or overloaded)
  3. FILTER: remove items that are not agent-actionable (external block, no actionable steps in 2–4h, done block).

Output: Constraint Brief draft (top-3 + classification + signals; for --layer=platform also Locus).

Phase 4 — Choose TOC tool

Input: Constraint Brief draft.

Action — Decision tree:

[All --layer values]

Weakest stage is obvious and data is available?
  → Five Steps (linear)

≥2 policy markers OR cognitive-freeze signal is high?
  → Five Steps + EC

2 candidates are equal, no clear "weaker" one?
  → EC (core conflict)

"Exploit" intuitively means "hire/buy" (instinct toward Elevate)?
  → EC first (marker: policy-constraint)

Solution is obvious but side-effects are a risk?
  → Five Steps + NBR

[--layer=platform only]

Locus = Subsystem is missing (no WP exists for the required SC)?
  → Five Steps + WP Gate Ritual (a new WP to create the subsystem is needed)

Locus = Handover between 2 subsystems?
  → EC required: D = "fix subsystem A" vs D' = "fix subsystem B"
  → + Coupling-analysis SOTA.011 at --depth=3: can the handover be eliminated entirely?

Locus = Role(B3) overloaded?
  → EC: D = "hire" vs D' = "automate the role"
  → NBR required (risk of heavy-coordination overhead)

All subsystems are stable but cross-system SC is not fulfilled?
  → Goldratt S&T Tree: missing assumptions (no SC, promise not recorded)

If EC: Quality gate ≥2 assumptions per side:

A (Common Goal): ...
B (Need): ...      C (Need): ...
D (Want): ...      D' (Want): ...
Assumptions D→B: ...
Assumptions D'→C: ...
Injection: ... (in assumption, not in action)

Empty / abstract assumptions → retry or fall back to Five Steps. EC applies to actions only, not to values.

NBR (required after injection): 3 negative branches + trim:

1. [risk 1] → trim: [how to neutralize]
2. [risk 2] → trim: [how to neutralize]
3. [risk 3] → trim: [how to neutralize]

Output: TOC tool trace (tool + EC details if applied + NBR trims).

Phase 5 — Compose stage map

Input: Constraint Brief + TOC tool trace + dependency list (including external).

Action:

  1. Identify the stages of reaching the target state (removing the constraint):
    • A stage = a system state, not an action ("≥5 pilots activated" ≠ "activate pilots")
    • Inside each node — parallel work items following the strict 4-category template (see below)
    • Between nodes — hard dependency ("otherwise we cannot proceed", not "would be nice")
  2. Inside each stage, distribute parallel work items across the 4 canonical TOC/Dettmer categories (always show all 4; empty = "—"):
    • Technical (Platform/Technical) — code, infrastructure, architecture, integrations, CI/CD, data pipelines
    • Process (Organizational/Process) — processes, communications, coordination between roles, user testing, onboarding
    • Policy (Policy/Governance) — standards, architecture decisions, approvals, policy changes
    • Cognitive (Cognitive/Capability) — training, documentation, mental models, building distinctions
  3. Identify external dependencies — work in other WPs, repos, or from external suppliers (separate block; always show)
  4. For --layer=platform — parallel Locus axis for each stage: Subsystem / Handover / Role(B3) — who specifically changes
  5. Build the graph (DP.WP.016 format): nodes = stages, edges = hard_dependency, external edges explicit
  6. Topological sort is required (no cycles)

Strict template principle: all 4 categories are always shown, even when empty ("—"). Analogy: an aircraft safety checklist — the "check landing gear" item remains even for a seaplane, so it is visible that the check was not skipped. The strict template prioritizes audit reliability over brevity.

Output: Stage Dependency Map (DP.WP.016 format) — in chat as markdown + optionally in ${IWE_GOVERNANCE_REPO:-DS-strategy}/inbox/WP-NNN/stage-map.md:

## Stage Dependency Map: <target> (layer: intra | platform)

### Stage 1: <noun-artifact label>
**Goal:** <system state after this stage>

**Work categories (strict template — always 4):**

**1. Technical (Platform):**
- work item A
- work item B (WP-NNN)

**2. Process:**
- —

**3. Policy:**
- work item C → requires /archgate

**4. Cognitive:**
- —

**External dependencies:** <target in another repo | none>

[For --layer=platform only — additional axis:]

**Locus:**
- Subsystem: <subsystem X> — change within a single system
- Handover: <A↔B> — change in the gap between systems
- Role(B3): <role> — change in the team

### → Stage 2: <label>
> Can start only after Stage 1 is complete (reason: <reason>)
...

Output (full artifact, printed to chat)

## System Card
**System type:** learning_pipeline | work_pipeline_intra | work_pipeline_platform | cohort_pipeline
**Target:** <target-ref>
**Layer:** intra | platform
**Structure:** <directions A–I | phases | C2 subsystems + handovers>
**Subsystems in scope:** <N — only when layer=platform>
**Data freshness:** <date + flags>

## SC-status map (Phase 2 SC-first)
[--layer=intra:]
| SC | Promise | Status | Signal |
| --- | --- | --- | --- |
| DP.SC.NNN | ... | ✅/⚠️/❌ | ... |

[--layer=platform — cross-system path:]
| SC | Cross-system path | Status | Weak link |
| --- | --- | --- | --- |
| DP.SC.020 | A → B → C → D | ⚠️ | handover B↔C |

## Constraint Brief
**Bottleneck:** <block / direction / phase / subsystem / handover / role>
**Trichotomy:** Work Flow | Work Process | Work Execution
**Class:** Policy | Resource | Cognitive
**Locus:** Subsystem | Handover | Role(B3)  [only when layer=platform]
**Affected SC:** [DP.SC.NNN, ...]

### Signal-scan
- Queue accumulation: <Δ over 7d or "no data">
- Flow Efficiency: <% or "no data">
- Dependency fan-in: <N handovers>
- Policy markers: <keywords or "none">
- Buffer: <status or "no data">
- Cognitive freeze: <indicators or "none">
- Handover count (cross-system): <N per user journey>          [only when layer=platform]
- Coupling tightness (SOTA.011): low | medium | high           [only when layer=platform]

## TOC tool: Five Steps | EC | Five Steps + EC | + Coupling-analysis

[If EC:]
### EC — core conflict
**A:** ...    **B:** ...    **C:** ...    **D:** ...    **D':** ...
**Assumptions D→B:** ...
**Assumptions D'→C:** ...
**Injection:** ...

### NBR
1. <risk> → trim: ...
2. <risk> → trim: ...
3. <risk> → trim: ...

## Stage Dependency Map (DP.WP.016)

### Stage 1: <label>
**Goal:** ...

**Work categories (strict template — always 4):**

**1. Technical (Platform):** ...
**2. Process:** ...
**3. Policy:** ...
**4. Cognitive:** ...

**External dependencies:** ...

**Locus:** Subsystem | Handover | Role(B3)  [only when layer=platform]

### → Stage 2: <label>
> Can start only after Stage 1 is complete (reason: ...)
...

## Alternatives (P2/P3 candidates)
- ...

## Skipped (not agent-actionable)
- ...

**First-stage budget: ~Nh**

---
Proceed with Stage 1 "<label>" or start with alternative P2 "<...>"?

Anti-patterns

  1. WP scan without SC-first — starting from a list of pending WPs without checking functional promises. Risk: mistaking the delivery channel for the product.
  2. Stage Map with dates/hours — the format is structural, not temporal. Dates → premature commitment.
  3. Soft dependencies in edges — only hard_dependency. "Would be nice to do first" means they belong in the same stage or are independent.
  4. Skip Subordinate → Elevate ("hire more people" as the first conclusion) — exploit capacity first.
  5. Policy mis-classified as Resource — hiring when the rule needs to change. Apply the discriminator from Schragenheim.
  6. Moving Herbie without a policy diagnosis — if the bottleneck has migrated 2+ times → it is policy-driven, not resource-driven.
  7. EC with empty assumptions — retry or fall back to Five Steps.
  8. Five Steps on cognitive freeze — do not speed up the flow; simplify the challenge or build the Skill.
  9. Output >1 screen for Constraint Brief — compress; do not expand the full tree (Stage Map may be longer).
  10. EC on values (D/D' = "flexibility" vs "stability") — reformulate as actions.
  11. Stage Dependency Map without external dependencies — if a stage genuinely depends on work in another WP, add an explicit external edge.
  12. Stage Map with an empty category and no "—" (soft template) — it is impossible to distinguish "forgot to think about it" from "decided to skip it". Strict template: always 4 categories; empty = "—".
  13. [platform only] Reducing the platform to a single system — picking one C2 subsystem as the bottleneck without checking cross-system SC. Symptom: answer "WP-117 = bottleneck" with no explanation of which SC it closes.
  14. [platform only] B3-blindness — a bottleneck in a role (bus-factor individual) is missed because roles are not in MAP.002. Force: explicitly check B3 in Phase 3.
  15. [platform only] Handover-blindness — both subsystems work, but SC breaks between them. Force: handover = a separate unit of analysis in Locus.
  16. [platform only] "Hire a manager" as the first Elevate for a Role-constraint — Role(B3) is resolved through role automation (subsystem), not hiring. Force: EC is required for a Role-constraint.
  17. [platform only] Coupling debt accumulation — the Subordinate plan does not account for coupling debt after the fix (SOTA.011). Force: --depth=3 with Coupling Model for a Handover-constraint.

Calibration (after each run)

Save to ${IWE_GOVERNANCE_REPO:-DS-strategy}/inbox/bottleneck-pick-runs/<YYYY-MM-DD>-<target>.yaml:

date: YYYY-MM-DD
target: <target-ref>
layer: intra | platform
system_type: learning_pipeline | work_pipeline_intra | work_pipeline_platform | cohort_pipeline
subsystems_in_scope: N                    # only when layer=platform
sc_failing: [DP.SC.NNN, ...]              # from Phase 2
identified_constraint:
  location: <block | subsystem | handover | role>
  trichotomy: work_flow | work_process | work_execution
  class: policy | resource | cognitive
  locus: subsystem | handover | role_b3   # only when layer=platform
instruments_used: [five_steps, ec, nbr, coupling_analysis]
stage_dependency_map:
  num_stages: N
  has_external_deps: true | false
  categories_filled: [platform, process, policy, cognitive]  # which of the 4 are non-empty
action_taken: <what was done>
was_correct: null          # filled in on Week Close
observed_throughput_change_after_2w: null
bottleneck_shifted_to: null                # filled in after 2 weeks

After 5–10 runs → analyze precision on Week Close, recalibrate signal weights.


Integration

  • Bottleneck = architecture decision (Policy type with alternatives) → propose /archgate
  • Stage Map contains an open-loop work item ≥3h → propose the Decomposer (/decompose) to break it down into physical Artifacts
  • Bottleneck = no covering WP → propose WP Gate ritual
  • Bottleneck-shift detection (re-run against the same target after a fix) → DP.M.061 layer reassessment (tech/operational/usage/behavioral)

Related documents (Pack)

  • DP.ROLE.054 — Constraint Analyst (methodology carrier)
  • DP.SC.045 — Constraint Analysis promise (what is delivered to the consumer)
  • DP.WP.016 — Stage Dependency Map format
  • DP.M.061 — Bottleneck-Shift Detection (supporting method after constraint removal)
  • PD.PRINC.046 — Mental Model as Constraint (Tendon)

Staging

STAGING.md: S-42, status: testing Promotion criteria: ≥2 smoke tests (passed) + ≥1 calibration session + pilot accepted ≥3 decisions without redirect + Pack Artifacts formalized (Phase 11 complete).

<!-- USER-SPACE --><!-- /USER-SPACE -->

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.