agentleFS
davidmatousekdavidmatousek/tachi/.claude/skills/~aod-discover/SKILL.md

~aod-discover

Skill95 starsChanged 37 days agoApache-2.0View on GitHub

Unified discovery skill with 4 entry points: /aod.discover (full flow: capture + score + validate), /aod.discover --seed (fast-track pre-vetted ideas with auto defaults), /aod.idea (capture + score only), /aod.validate (PM validation for existing idea). Use this skill when you need to capture ideas, run discovery, validate ideas with PM, generate user stories, log feature requests, or add items to the ideas backlog.

The file

Show its rules and text

We split this skill into the rules it gives an agent

30 rules. About 0 are common, 12 are distinctive, and 18 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈12

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 18

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If invoked via /aod.discover --seed → run Seed Mode (Steps 1, 2, 5a, 5b, 8-seed)

    line 20
  2. Very specific:

    If invoked via /aod.discover → run Full Flow (Steps 1-8)

    line 21
  3. Very specific:

    If invoked via /aod.idea → run Capture Only (Steps 1-5)

    line 22
  4. Very specific:

    If invoked via /aod.validate → run Validate Only (Steps 6-8)

    line 23
  5. Very specific:

    Check if arguments contain --seed flag.

    line 27
  6. Very specific:

    Remove --seed from the idea description text

    line 28
  7. Very specific:

    MUST: use .aod/scripts/bash/create-issue.sh for ALL GitHub Issue creation.

    line 34
  8. Very specific:

    File reference pattern: If arguments contain read #### F-NNN: {title} in {file_path}, this is a directive to read the full feature section …

    line 60
  9. Distinctive:

    Use the extracted content as the idea description (title = heading, body = full epic block including Goal, Stories, Interface Contract, and …

    line 65
  10. Distinctive:

    No scanning, parsing, or incrementing is needed — GitHub handles ID assignment

    line 77
  11. Distinctive:

    Use AskUserQuestion to determine the idea source:

    line 83
  12. Distinctive:

    After ICE scoring, prompt the user for evidence supporting this idea.

    line 125
  13. Distinctive:

    Create GitHub Issue (Step 5a) and regenerate BACKLOG.md (Step 5b), then STOP.

    line 158
  14. Distinctive:

    Create a GitHub Issue for lifecycle tracking:

    line 200
  15. Very specific:

    MUST: use the standalone create-issue.sh script (do NOT call gh issue create directly — the script handles both issue creation and project …

    line 226
  16. Very specific:

    If gh is unavailable, skip silently (graceful degradation).

    line 236
  17. Very specific:

    After GitHub Issue creation, run .aod/scripts/bash/backlog-regenerate.sh to update the backlog snapshot.

    line 240
  18. Very specific:

    Before proceeding to PM validation in the Full Flow (/aod.discover), check the governance tier:

    line 244
  19. Very specific:

    If invalid or missing, display: Usage: /aod.validate NNN (or #NNN or IDEA-NNN)

    line 268
  20. Distinctive:

    Read the issue body to extract idea details (description, ICE score, evidence, source, status).

    line 286
  21. Very specific:

    Use the Task tool with product-manager subagent_type:

    line 306
  22. Very specific:

    If PM returns APPROVED or APPROVED_WITH_CONCERNS: Continue to Step 7.

    line 374
  23. Distinctive:

    Default persona: "Template Adopter" if not evident from idea.

    line 379
  24. Distinctive:

    Present for confirmation: Use AskUserQuestion:

    line 380
  25. Very specific:

    Update the GitHub Issue body to include the user story in a new ## User Story section

    line 396
  26. Very specific:

    Update Status: in the body to "Validated"

    line 397
  27. Distinctive:

    Add a comment documenting PM approval with rationale and evidence quality

    line 398
  28. Distinctive:

    If idea was auto-deferred and PM approved, add PM override note as a comment:

    line 399
  29. Very specific:

    Step 5a (GitHub Issue Creation): Create issue with stage:discover label using create-issue.sh, same structured body format.

    line 499
  30. Distinctive:

    PM approves auto-deferred idea: Document override rationale as GitHub Issue comment

    line 541

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

AOD Discover Skill

Purpose

Unified discovery skill for the AOD Lifecycle's Discover stage. Handles idea capture, ICE scoring, evidence gathering, and PM validation with 3 entry points:

  • /aod.discover — Full flow: capture idea + ICE score + evidence + PM validation + user story
  • /aod.discover --seed — Seed mode: fast-track pre-vetted ideas (auto ICE, skip prompts, skip PM)
  • /aod.idea — Capture only: capture idea + ICE score + evidence (no PM validation)
  • /aod.validate — Validate only: submit existing idea for PM review + user story generation

Entry Point Detection

Determine the entry point from the invoking command:

  • If invoked via /aod.discover --seed → run Seed Mode (Steps 1, 2, 5a, 5b, 8-seed)
  • If invoked via /aod.discover → run Full Flow (Steps 1-8)
  • If invoked via /aod.idea → run Capture Only (Steps 1-5)
  • If invoked via /aod.validate → run Validate Only (Steps 6-8)

--seed Flag Detection

Check if arguments contain --seed flag. If present:

  • Remove --seed from the idea description text
  • Set entry point to Seed Mode
  • Seed mode is designed for pre-vetted ideas (e.g., from consumer guides) that don't need scoring or validation

Critical Constraints

MUST use .aod/scripts/bash/create-issue.sh for ALL GitHub Issue creation. NEVER call gh issue create directly.

Why: Direct gh issue create bypasses the board sync pipeline, causing issues to silently disappear from the project board. The create-issue.sh script delegates to aod_gh_create_issue() which includes automatic board placement via aod_gh_add_to_board(). See KB Entry #20 for root cause analysis.

Prohibited pattern (DO NOT use):

# WRONG — bypasses board sync
gh issue create --title "..." --label "stage:discover" --body "..."

Required pattern:

# CORRECT — includes board sync automatically
bash .aod/scripts/bash/create-issue.sh --title "Feature title" --body "Markdown body" --stage discover

Source of Truth

GitHub Issues are the sole source of truth for backlog items. All idea state (ICE scores, status, evidence) is stored in the GitHub Issue body. BACKLOG.md is an auto-generated view regenerated from Issues.


Step 1: Parse Input (Capture + Full Flow)

Extract the idea description from user arguments. If no description is provided, ask the user to describe their idea.

File reference pattern: If arguments contain read #### F-NNN: {title} in {file_path}, this is a directive to read the full feature section from the referenced consumer guide:

  1. Open {file_path} using the Read tool
  2. Locate the heading #### F-NNN: {title}
  3. Extract everything from that heading down to (but not including) the next --- separator
  4. Use the extracted content as the idea description (title = heading, body = full epic block including Goal, Stories, Interface Contract, and DoD)

This pattern is generated by /aod.kickstart in the Quick Start Commands section of consumer guides. It enables one-command-per-feature GitHub Issue creation with the complete epic content.

Preserve full input: If the user provides detailed content beyond a summary (e.g., stories, acceptance criteria, interface contracts, Definition of Done), capture ALL of it verbatim. This full content is stored in the GitHub Issue body in Step 5a so downstream commands (/aod.define, /aod.spec) can build on it.

Step 2: ID Assignment

The GitHub Issue number is the canonical idea ID. There is no separate IDEA-NNN numbering scheme.

  • The ID is assigned automatically when create-issue.sh runs in Step 5a
  • Refer to ideas by their GitHub Issue number using #NNN format (e.g., #42)
  • No scanning, parsing, or incrementing is needed — GitHub handles ID assignment

Step 3: Capture Source and ICE Score

Source

Use AskUserQuestion to determine the idea source:

Question: "Where did this idea come from?"
Options:
  - Brainstorm: "Generated during a brainstorming or planning session"
  - Customer Feedback: "Reported by a customer or based on user research"
  - Team Idea: "Suggested by a team member during development"
  - User Request: "Directly requested by a user or stakeholder"

ICE Scoring

Score all three ICE dimensions in one batched AskUserQuestion call — a single call carrying three questions (Impact, Confidence, Effort) — so the user answers them together instead of in three sequential round-trips:

One AskUserQuestion call, three questions:

Q1 — Impact: "How much value does this deliver to users?"
  - High (9): "Transformative — significant user value"
  - Medium (6): "Solid improvement — meaningful but incremental"
  - Low (3): "Minor enhancement — small quality-of-life fix"

Q2 — Confidence: "How sure are we this will succeed?"
  - High (9): "Proven pattern — strong evidence it will work"
  - Medium (6): "Some unknowns — reasonable confidence with gaps"
  - Low (3): "Speculative — significant uncertainty"

Q3 — Effort (Ease of Implementation): "How easy is this to build?"
  - High (9): "Days of work — straightforward implementation"
  - Medium (6): "Weeks of work — moderate complexity"
  - Low (3): "Months of work — significant engineering effort"

Allow "Other" for custom numeric values (1-10) on each question.

Compute: ICE Total = Impact + Confidence + Effort (range 3-30)

Step 4: Evidence Prompt (Capture + Full Flow)

After ICE scoring, prompt the user for evidence supporting this idea.

Use AskUserQuestion:

Question: "Who has this problem, and how do you know?"
Header: "Evidence"
Options:
  - Customer Feedback: "Based on direct customer reports, support tickets, or user research"
  - Team Observation: "Noticed by the team during development, testing, or internal usage"
  - Analytics Data: "Supported by usage metrics, error rates, or behavioral analytics"
  - User Request: "Explicitly requested by a user, stakeholder, or partner"

Allow "Other" for free-text evidence statements.

Store the response as the evidence field. This value is used in:

  • Step 5a: GitHub Issue body (## Evidence section)
  • Step 6: PM validation (evidence quality evaluation)
  • BACKLOG.md Discover section (Evidence column)

If the user selects a predefined category: Store as "{Category}: {any additional detail}" if they provide detail, or just "{Category}" if not.

If the user provides free text via "Other": Store the full text as-is.

Step 5: Apply Auto-Defer Gate

Determine status based on ICE total:

  • Total < 12: Set status to "Deferred" (auto-deferred)
  • Total >= 12: Set status to "Scoring"

If auto-deferred AND entry point is Full Flow (/aod.discover):

  • Create GitHub Issue (Step 5a) and regenerate BACKLOG.md (Step 5b), then STOP. Do not proceed to PM validation.
  • Display result with guidance:
IDEA CAPTURED — AUTO-DEFERRED

ID: #{NNN}
Idea: {description}
Source: {source}
Evidence: {evidence}
ICE Score: {total} (I:{impact} C:{confidence} E:{effort})
Priority Tier: Deferred
Status: Deferred

This idea was auto-deferred (score < 12).
To request PM override: `/aod.validate #{NNN}`
To re-score with new information: `/aod.score #{NNN}`

If auto-deferred AND entry point is Capture Only (/aod.idea): Create GitHub Issue, display same result, exit.

If NOT auto-deferred AND entry point is Capture Only (/aod.idea): Create GitHub Issue, display result and exit:

IDEA CAPTURED

ID: #{NNN}
Idea: {description}
Source: {source}
Evidence: {evidence}
Date: {YYYY-MM-DD}
ICE Score: {total} (I:{impact} C:{confidence} E:{effort})
Priority Tier: {tier}
Status: Scoring

Next: Run `/aod.validate #{NNN}` to submit for PM review, or continue capturing ideas with `/aod.idea`.

If NOT auto-deferred AND entry point is Full Flow (/aod.discover): Continue to Step 5a, then Step 5c.

Step 5a: GitHub Issue Creation (All Entry Points)

Create a GitHub Issue for lifecycle tracking:

  1. Build the issue body using the structured format. If the user provided detailed content (stories, acceptance criteria, interface contracts, DoD), include it in the ## Detail section:
    ## Idea
    
    {idea_summary — first sentence or goal line}
    
    ## Detail
    
    {full user-provided content: stories, acceptance criteria, interface contracts, DoD, etc.
    Include verbatim — do not summarize or omit. Omit this section if user only provided a summary.}
    
    ## ICE Score
    Impact: {impact}, Confidence: {confidence}, Effort: {effort} = **{total}**
    
    ## Evidence
    {evidence}
    
    ## Metadata
    - Source: {source}
    - Priority: {priority_tier}
    - Date: {YYYY-MM-DD}
    - Status: {status}
  2. MUST use the standalone create-issue.sh script (do NOT call gh issue create directly — the script handles both issue creation and project board sync):
    bash .aod/scripts/bash/create-issue.sh \
      --title "{idea_description}" \
      --body "$BODY" \
      --stage discover \
      --type idea
    The script outputs the issue number on stdout. Capture it — this becomes the canonical idea ID (#NNN).
  3. If gh is unavailable, skip silently (graceful degradation).

Step 5b: Regenerate BACKLOG.md

After GitHub Issue creation, run .aod/scripts/bash/backlog-regenerate.sh to update the backlog snapshot. If gh is unavailable, skip silently.

Step 5c: Check Governance Tier (Full Flow only)

Before proceeding to PM validation in the Full Flow (/aod.discover), check the governance tier:

  1. Read .aod/memory/constitution.md
  2. Find the governance tier configuration (governance: → tier: value)
  3. Valid values: light, standard, full. Default: standard

If tier is light:

  • PM validation is optional — skip Steps 6-7 in the Full Flow
  • Display note: "Note: Light governance tier — PM validation skipped. Run /aod.validate #NNN to manually request PM review."
  • Exit. The user can still manually invoke /aod.validate #NNN at any time

If tier is standard or full: Continue to Step 6 as normal.

Note: The /aod.validate entry point always runs PM validation regardless of tier — it is an explicit user request for PM review.

Step 6: PM Validation (Validate + Full Flow)

For /aod.validate entry point: Parse and find idea

Extract the idea identifier from user arguments. Accept three formats:

  • NNN (bare number, e.g., 21): Direct GitHub Issue lookup
  • #NNN (hash-prefixed, e.g., #21): Strip # prefix, direct lookup
  • IDEA-NNN (legacy, e.g., IDEA-009): Search issue titles for [IDEA-NNN] bracket tag

If invalid or missing, display: Usage: /aod.validate NNN (or #NNN or IDEA-NNN)

Find the matching GitHub Issue:

For numeric input (NNN or #NNN):

source .aod/scripts/bash/github-lifecycle.sh && aod_gh_find_issue NNN

For legacy IDEA-NNN input:

source .aod/scripts/bash/github-lifecycle.sh && aod_gh_find_issue "[IDEA-NNN]"

Error conditions:

  • Issue not found: "Error: No GitHub Issue found for {identifier}"
  • Issue already has stage:define or later: "Error: #{issue_number} has already progressed past discovery"

Read the issue body to extract idea details (description, ICE score, evidence, source, status).

Display idea for review:

IDEA FOR PM VALIDATION

ID: #{issue_number}
Idea: {description}
Source: {source}
Evidence: {evidence}
Date: {date}
Status: {status}
ICE Score: {total} (I:{impact} C:{confidence} E:{effort})
Priority Tier: {tier}
Auto-Deferred: {Yes if status is Deferred, otherwise No}

Launch PM Review

Use the Task tool with product-manager subagent_type:

Review this idea for product backlog inclusion. You are the product-manager
guardian of product-spec alignment — apply your §5 Quality Standards rubric.

Idea: {idea_description}
ICE Score: {total} (I:{impact} C:{confidence} E:{effort})
Priority Tier: {tier}
Current Status: {status}
Auto-Deferred: {yes/no}
Evidence: {evidence}

Ground your judgement in the actual product artifacts — read what is relevant:
- Vision:       docs/product/01_Product_Vision/product-vision.md
- Roadmap:      docs/product/03_Product_Roadmap/
- OKRs:         docs/product/06_OKRs/
- User Stories: docs/product/05_User_Stories/README.md

Evaluate against these criteria:
1. Vision & positioning fit — does the idea serve the product vision and target users?
2. Strategic alignment — does it fit the roadmap timeline and support a current-quarter OKR?
3. User value — is there a clear, user-focused problem and meaningful benefit?
4. ICE reasonableness — are Impact/Confidence/Effort credible given the description and evidence?
5. Evidence sufficiency — is the evidence specific enough to justify pursuing this now?
6. Backlog-readiness — is scope bounded enough to become a well-formed user story?

If evidence is empty or "No evidence provided", flag it explicitly:
"No evidence provided — recommend gathering evidence before proceeding."
You may still approve on strong merit, but note the evidence gap in your rationale.

If auto-deferred (score < 12), justify why this idea should or should not override
the auto-defer gate.

Apply a thinking lens when warranted (Pre-Mortem for risky/unclear value,
First Principles to challenge inherited assumptions); fold any lens findings into
the rationale.

Respond with:
STATUS: [APPROVED | APPROVED_WITH_CONCERNS | CHANGES_REQUESTED | BLOCKED]
RATIONALE: [2-4 sentences — vision/roadmap alignment, evidence quality in prose, and user value]

Handle Rejection

Map the PM STATUS: verdict to an action. No value falls through — a missing or unparseable STATUS is treated as CHANGES_REQUESTED (fail safe toward human re-review, never a silent capture):

STATUS: verdictAction
APPROVEDProceed to Step 7 (capture as a validated idea).
APPROVED_WITH_CONCERNSProceed to Step 7, AND record the PM's concerns (from RATIONALE:) as a comment on the GitHub Issue so they travel with the story.
CHANGES_REQUESTEDDo NOT capture. Route back to the human for refinement: surface the PM RATIONALE:, set the Issue Status: to "Changes Requested", and stop.
BLOCKEDHard stop — no capture. Route to the human with the PM RATIONALE:; set the Issue Status: to "Blocked". Never auto-override a BLOCKED verdict.

For a CHANGES_REQUESTED or BLOCKED verdict, update the GitHub Issue (add the PM rationale as a comment, set the Status: line accordingly), display the result, then exit:

PM VALIDATION: {status}

ID: #{issue_number}
Idea: {description}
PM Rationale: {rationale}

The idea was not captured ({status}). To re-submit: re-score with `/aod.score #{issue_number}`, then run `/aod.validate #{issue_number}` again.

Handle Approval

If PM returns APPROVED or APPROVED_WITH_CONCERNS: Continue to Step 7. For APPROVED_WITH_CONCERNS, the concerns recorded above travel with the user story.

Step 7: User Story Generation

  1. Generate user story: Transform idea into "As a [persona], I want [action], so that [benefit]" format.
    • Default persona: "Template Adopter" if not evident from idea.
  2. Present for confirmation: Use AskUserQuestion:
    Options:
      - Accept: "Save this user story as-is"
      - Edit: "Let me modify the user story text"
  3. Determine priority: Map ICE score to priority rank:
    • P0 (25-30): Priority = 1
    • P1 (18-24): Priority = 2
    • P2 (12-17): Priority = 3
    • Deferred (<12, PM override): Priority = 4

Step 8: Update GitHub Issue and Report

Update GitHub Issue

  1. Update the GitHub Issue body to include the user story in a new ## User Story section
  2. Update Status: in the body to "Validated"
  3. Add a comment documenting PM approval with rationale and evidence quality
  4. If idea was auto-deferred and PM approved, add PM override note as a comment:
    **PM Override**: #{issue_number} was auto-deferred (ICE score {total} < 12) but approved by PM. Rationale: {pm_rationale}

Regenerate BACKLOG.md

Run .aod/scripts/bash/backlog-regenerate.sh only for the /aod.validate entry point (which skips capture). In the full /aod.discover flow, Step 5b already regenerated the snapshot at capture — do NOT regenerate again here (the user-story update does not change the stage-keyed backlog). This keeps every exit path — capture-only, deferred, full-flow-approved, and /aod.validate — regenerating BACKLOG.md exactly once.

Report result

For /aod.validate entry point:

PM VALIDATION: APPROVED

ID: #{issue_number}
Idea: {description}
PM Rationale: {rationale}

User Story Created:
  Story: "{user_story_text}"
  Priority: {priority}
  Status: Validated

GitHub Issue #{issue_number} updated with user story and PM approval.

Next: Run `/aod.define {topic}` to create a PRD from this user story.

For /aod.discover entry point (full flow):

AOD DISCOVERY COMPLETE

Idea Captured:
  ID: #{issue_number}
  Description: {description}
  Source: {source}
  Evidence: {evidence}

ICE Scoring:
  Score: {total} (I:{impact} C:{confidence} E:{effort})
  Priority Tier: {tier}

PM Validation: APPROVED
  Rationale: {rationale}

User Story Created:
  Story: "{user_story_text}"
  Priority: {priority}
  Status: Validated

Next: Run `/aod.define {topic}` to create a PRD from this user story.

ICE Scoring Reference

Quick-Assessment Anchors

DimensionHigh (9)Medium (6)Low (3)
ImpactTransformativeSolid improvementMinor enhancement
ConfidenceProven patternSome unknownsSpeculative
Effort (Ease)Days of workWeeks of workMonths of work

Priority Tiers

Score RangePriorityAction
25-30P0 (Critical)Fast-track to development
18-24P1 (High)Queue for next loop
12-17P2 (Medium)Consider when capacity allows
< 12DeferredAuto-defer; requires PM override via /aod.validate

Auto-Defer Gate

Ideas scoring below 12 are automatically deferred. In the full flow (/aod.discover), the flow stops and no PM validation occurs. Use /aod.validate #NNN to request PM override for deferred ideas.


Seed Mode (--seed Flag)

Seed mode is a fast-track path for pre-vetted ideas (e.g., from a consumer guide generated by /aod.kickstart). It skips all interactive prompts, auto-assigns sensible defaults, and creates the GitHub Issue immediately.

When to Use

  • Importing seed stories from a consumer guide
  • Batch-loading pre-planned features into the backlog
  • Any idea where scoring and validation are unnecessary because the idea is already committed

Seed Mode Flow

  1. Step 1 (Parse Input): Extract idea description (strip --seed from arguments). If empty, ask for description.
  2. Step 2 (ID Assignment): Same as normal — GitHub Issue number is canonical ID.
  3. Auto-assign defaults (skip Steps 3-4):
    • Source: Consumer Guide
    • ICE Score: Impact=8, Confidence=7, Effort=7 (Total=22, P1 High)
    • Evidence: Seed story from consumer guide — pre-vetted during kickstart planning
  4. Skip auto-defer gate (Step 5): Score is 22, always passes.
  5. Step 5a (GitHub Issue Creation): Create issue with stage:discover label using create-issue.sh, same structured body format.
  6. Step 5b (Regenerate BACKLOG.md): Same as normal.
  7. Skip PM validation (Steps 5c-7): No governance check, no PM review, no user story generation.
  8. Report result:
SEED IDEA CAPTURED

ID: #{issue_number}
Idea: {description}
Source: Consumer Guide
ICE Score: 22 (I:8 C:7 E:7)
Priority Tier: P1 (High)
Status: Scoring

Seed mode — ICE scoring, evidence, and PM validation skipped.
Next: Run `/aod.define {topic}` to create a PRD, or continue seeding with `/aod.discover --seed`.

Seed Mode Defaults

FieldDefault ValueRationale
SourceConsumer GuidePre-vetted during kickstart
Impact8High — guide features are user-facing
Confidence7Good — planned but not yet validated
Effort7Moderate — typical feature scope
EvidenceSeed story from consumer guide — pre-vetted during kickstart planningAcknowledges provenance
StatusScoringReady for define stage

Edge Cases

  • No description provided: Prompt user for idea description
  • Custom ICE score outside 1-10: Clamp to valid range (1 minimum, 10 maximum)
  • Duplicate idea text: Allow it — GitHub assigns a unique issue number automatically
  • Issue not found (validate): Search GitHub Issues by number, display error if not found
  • Already Rejected (validate): Cannot re-validate directly — user must re-score first with /aod.score
  • Already past discovery (validate): Cannot re-validate — idea has progressed
  • PM timeout or error: Report the error and suggest retrying
  • User edits story to empty string: Re-prompt for user story text
  • PM approves auto-deferred idea: Document override rationale as GitHub Issue comment
  • Auto-deferred in full flow: Flow stops — no PM validation, no user story
  • gh unavailable: Skip GitHub Issue creation silently (graceful degradation)
  • --seed with no description: Prompt for idea description (flag alone is not enough)

Quality Checklist

  • GitHub Issue created via create-issue.sh (NOT gh issue create directly) with structured body, stage:discover label, and type:idea label
  • Entry point correctly detected (discover/discover --seed/idea/validate)
  • --seed flag: stripped from description, defaults applied, prompts skipped
  • Idea ID is the GitHub Issue number (#NNN), assigned by create-issue.sh
  • Source captured from user selection
  • Evidence prompted after ICE scoring (predefined categories + free text)
  • Evidence included in GitHub Issue body and display outputs
  • ICE score computed correctly (additive I+C+E)
  • Auto-defer gate applied (< 12 = Deferred, flow stops in full flow)
  • Issue added to Projects board with correct Status column (verified via script output)
  • BACKLOG.md regenerated after Issue creation
  • Governance tier read from constitution (light/standard/full, default: standard)
  • Light tier: PM validation skipped in full flow, with note to user
  • PM validation evaluates evidence quality as free-text in the PM RATIONALE: (no separate enum)
  • PM validation invoked via Task tool (product-manager subagent)
  • Rejection: GitHub Issue updated, flow exits
  • Approval: User story generated in proper format
  • User story presented for confirmation
  • GitHub Issue updated with user story and PM approval
  • PM override documented if idea was auto-deferred
  • BACKLOG.md regenerated after validation
  • Result reported with next step guidance

Common Rationalizations

RationalizationReality
"I'll just use gh issue create directly — same result"Critical Constraints (line 34) require create-issue.sh; direct gh bypasses board sync per the line 38 warning.
"Score 11 is close enough — I'll skip the auto-defer gate"Step 5 (line 162) hard-codes < 12 as Deferred and stops the full flow before PM review.
"Evidence is optional, I'll mark it 'No evidence provided'"Step 6 PM prompt (line 336) flags empty evidence; PM may approve but documents the gap in rationale.
"Seed mode skips ICE — I can use it for any quick idea"Seed Mode (line 481) is for pre-vetted consumer-guide stories only; defaults Impact=8, Confidence=7 per line 493.
"Light tier means PM validation is permanently disabled"Step 5c (line 261) skips PM in full flow only; line 267 confirms /aod.validate always runs PM regardless.

Red Flags

  • Agent calls gh issue create directly, bypassing aod_gh_add_to_board per line 36 KB Entry #20.
  • Agent advances an auto-deferred idea to PM validation without /aod.validate override per Step 5 (line 167).
  • Agent skips Step 5b BACKLOG.md regeneration after Issue creation, leaving the snapshot stale per line 248.
  • Agent assigns a separate IDEA-NNN ID instead of using the GitHub Issue number per Step 2 (line 73).
  • Agent ignores legacy IDEA-NNN format support in Step 6 (line 276) and rejects valid input.
  • Agent submits the user story without the AskUserQuestion confirmation step per Step 7 (line 375).

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.