agentleFS
ToddGBensonToddGBenson/keel/.claude/skills/backlog-prioritization/SKILL.md

backlog-prioritization

Skill0 starsChanged 3 months agoApache-2.0View on GitHub

Rank an open backlog and group the top of it into a coherent epic. Use when deciding what to build next, running /plan, grouping stories for batched verification, or when asked to prioritize issues. Covers ranking ideas from existing RICE scores and defects from their recorded severity (never a common scale), coherence over raw rank, and what makes a group verifiable together.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    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:

    Every item on the backlog arrived through an intake route, and each route records its own ranking input.

    line 21
  2. Distinctive:

    Prefer a group that shares a surface: the same component, the same data, the same user journey, the same trust boundary.

    line 95
  3. Distinctive:

    Every slate names what was not selected and why.

    line 114
  4. Distinctive:

    If asked how long the epic will take, the honest answer is the story count and their individual sizes, not a date derived from a velocity …

    line 127

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Backlog prioritization

Consumers: product-owner, delivery-lead, /plan, ADR-0002.

Prioritization is selection, and selection is a human decision

You produce a ranked slate with reasoning. A human picks. This is not deference for its own sake — deciding what to build is a scope judgement, and scope is the boundary the unattended runner is forbidden to cross (P2, docs/18). A command that ranked and committed would walk back through that boundary from the other side.

Present the recommendation confidently. Do not create the epic until told.

Rank from what you already have

Every item on the backlog arrived through an intake route, and each route records its own ranking input. Rank each item on the input its route recorded — never on one you made up. Re-scoring is where planning turns into ceremony, and a score nobody assessed is worse: it is a fabricated input dressed as a recorded decision (PD-7).

KindIntake routeRecorded inputPath
Idea / story/idea → G0RICE score on the issueRICE
Defectbug.ymlSeverity field from the bug form, plus its scope-of-impact and flagsSeverity
Security finding, compliance gap, EOLsecurity-finding / POA&M / choreSLA due date from docs/10-definitions.md § SeveritySLA

Ideas — RICE from G0

The backlog already carries a RICE score set at G0 (/idea). Use it.

Reach × Impact × Confidence / Effort

An idea with no RICE never passed G0. It is not eligible for planning — it is eligible for /idea. Say so in the slate.

Defects — recorded severity, not RICE

A bug filed through bug.yml never passes /idea, so it never has a RICE score, and it should not: G0 says risk-reduction work is not RICE-ranked against features. Never back-fill a RICE score for a defect. Rank defects on the severity recorded on the bug form:

  1. Severity — Blocker › Major › Minor (docs/10-definitions.md § Severity).
  2. Within a severity, the data loss or data exposed flag ranks first.
  3. Then the recorded scope of impact — more users, a core segment, longer-standing, first.

A defect with no recorded severity does not meet the bug Definition of Ready. Say so and route it back to triage; do not assign one yourself to make it rankable.

A defect ticked security-relevant is a security finding in disguise: it takes the SLA path, with severity set from realistic exploitability (scanner-triage), not the bug-form dropdown.

Security findings, compliance gaps, EOL — the SLA date

These bypass ranking (docs/02). They are listed in due-date order at the top of the slate as constraints the plan must fit around, not candidates competing with features.

No common scale across kinds

A RICE score and a defect severity measure different things. There is no common scale that makes "RICE 240" comparable to "Major, ~30 users" — and inventing a conversion factor is just a fabricated score one step removed. So:

  • Rank within each kind only, and present the kinds as separate lists.
  • State the path for every item in the slate, so nobody reads a mixed table as one ranking.
  • Interleaving across kinds is the human's call, made when they approve the slate. Your recommendation may say which you would take first and why — in words, not in a number.

An epic may still mix kinds when they share a surface (coherence, below); the per-item path column keeps that honest.

Adjustments, on every path

Then adjust for three things no recorded score can see:

AdjustmentWhy
Dependency orderA high-ranked item blocked by a low-ranked one is not actually next. Order beats score.
Risk flags (security-relevant, ai-relevant)These carry a mandatory G2 and cannot be batch-verified (ADR-0002). Real cost, not a tiebreak.
DecayAn issue open for months with no movement is either mis-scored or not actually wanted. Say which.

Coherence beats rank

The top N by score is usually the wrong epic. A group of unrelated high scorers is a list, not an epic, and it produces exactly the verification problem batching is meant to solve — five disconnected attack surfaces reviewed in one pass, with no seams worth looking at.

Prefer a group that shares a surface: the same component, the same data, the same user journey, the same trust boundary. Then the batched review can ask the question only a batched review can — what did these do to each other?

A slightly lower-ranked but coherent group beats a higher-ranked scatter. Say when you made that trade and why.

Sizing the group

Three to five stories. Below three, batching saves little over doing them individually. Above five, the combined surface stops fitting in one review and the pass degrades into a formality (ADR-0002 D5, enforced by scripts/check-batch-eligibility.sh).

Flagged stories do not count toward a batch because they are not in one — they are verified individually. If the coherent group is mostly flagged, that is a signal the whole group belongs in the full-ceremony lane, and saying so is the right answer.

The deferred list is not optional

Every slate names what was not selected and why. A plan that shows only the chosen work hides the decision, and the decision is the entire product of this exercise.

Three honest reasons to defer, and they are different: lower value · blocked by something · not coherent with this group, will lead the next one. The third is not a demotion, and labelling it as one loses information.

What this deliberately does not do

No dates. No velocity. No story points. No capacity. An epic is a grouping, not an iteration (ADR-0002 D6). Those metrics are excluded platform-wide because measuring them corrupts the estimates the gates depend on — see docs/00-overview.md § Anti-metrics.

If asked how long the epic will take, the honest answer is the story count and their individual sizes, not a date derived from a velocity that does not exist.

Output shape

## Recommended epic — <name>
<one line: the shared surface that makes these one group>

| # | Story | Path | Score | Flags | Why in this group |
|---|-------|------|-------|-------|-------------------|

Path is `RICE`, `Severity`, or `SLA`; Score is the recorded value for that path (a RICE number,
Blocker/Major/Minor, or a due date). Never a number converted from another path.

Batch-eligible: yes/no — <if no, which stories and why>

SLA-bound items (constraints, not candidates): <# — due date>, or "none"

## Deferred
| # | Story | Path | Reason (lower value / blocked / not coherent — leads next / not rankable — why) |

Then stop and ask for approval.

Controls

SA-3 (lifecycle), SA-4 (requirements), RA-3 (risk assessment), PM-11 (mission/business process), CM-3 (traceability — the epic is an issue, so parentage is auditable).

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.