agentleFS
dougcunhadougcunha/TokenHound/.agents/skills/sdd-triage/SKILL.md

sdd-triage

Skill3 starsChanged 36 days agoMITView on GitHub

SDD triage that recommends to the human the process level for a feature, fix, or refactoring request — full SDD, lean SDD, or spot change — from risk signals with evidence, before any artifact exists. Use when starting sdd-orchestrate-flow on a feature without a checkpoint or when asked whether a request is worth the SDD flow. Don't use for resuming a feature with a checkpoint or for slicing a broad request into several PRDs (use sdd-orchestrate-prds).

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    Gather signals: Read AGENTS.md for critical areas, risk skills, and commands.

    line 19
  2. Very specific:

    Output: every signal filled with evidence or not measured.

    line 32
  3. Very specific:

    Deliver: Commit only when asked, through the commit skill.

    line 52

≈ “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

Triage a request for SDD

Triage spends little to avoid spending a lot: it gathers signals with evidence, applies a rubric, and recommends a level; the human decides. Budget: up to ten targeted searches or reads, with no build or tests. Triage that needs more than that to understand the request already points to sdd-full.

LevelPath
sdd-fullsdd-orchestrate-flow with every gate
sdd-leansdd-orchestrate-flow with a short PRD (problem, RF with acceptance, out of scope), HIL 1 and HIL 2 merged into one decision, a plan of one or two tasks, and independent review kept
spotThe spot branch of this skill, outside SDD
  1. Understand. Extract the type (feature, fix, refactoring), the primary outcome, and the expected behavior. With an existing checkpoint for the request, stop and resume through sdd-orchestrate-flow. With more than one primary outcome, recommend sdd-full through sdd-orchestrate-prds and go to step 4. Output: request with one identified primary outcome, or a resume or slicing referral.
  2. Gather signals. Read AGENTS.md for critical areas, risk skills, and commands. Locate the likely change points through the repository's context graph when it has one (e.g. graft ask, graft callers <symbol> --depth 2), or by direct search. Fill each signal with path:line evidence or not measured:

    SignalQuestion
    S1 Public contractDoes it change what a consumer sends or receives: an API or MCP surface, a serialized DTO, a schema or database script, a file format, or configuration? Documentation-only text (description, example, grouping) counts as absent.
    S2 Critical areaDoes it touch an area that AGENTS.md declares critical or that requires a risk skill (e.g. credentials, provider authentication, payments)?
    S3 ConcurrencyDoes it involve a transaction, lock, idempotency, queue, background polling, or shared state?
    S4 Blast radiusHow many production files, modules, and transitive callers change behavior?
    S5 Open decisionIs there an ambiguous requirement or an unanswered product decision?
    S6 ObligationsHow many distinct observable behaviors does the request require?
    S7 TestsDo tests exercise the behavior, or does a direct regression test fit?
    S8 ReversibilityDoes it persist data, migrate state, or trigger an external action that is hard to undo?

    Output: every signal filled with evidence or not measured.

  3. Recommend. Apply the rubric, whose thresholds are a starting point calibrated by the triage log:
    • sdd-full when any holds: S1, S2 with a behavior change, S5, S8, S6 above three obligations, or S4 across more than one module.
    • spot when all hold: one module, up to three production files, none of S1, S2, S3, S5, or S8, an unambiguous expected behavior, and a feasible regression test (S7).
    • sdd-lean otherwise. A not measured signal that would decide the level counts as present. Output: rubric level with the deciding signals.
  4. HIL 0. Present the level the rubric recommends, the deciding signals with evidence, and what each level costs in artifacts, HILs, and review. Ask with the available question tool, recommended option first. Silence keeps the decision pending. With --mode auto, do not ask: the rubric level is the decision, recorded as autonomous, and the protocol in .agents/skills/sdd-orchestrate-flow/references/autonomous-mode.md applies from here on. Append one line to tasks/triage-log.jsonl with date, summarized request, signals, rubric level, and human decision, or autonomous in auto mode. Output: level decided by the human and recorded.
  5. Continue. sdd-full and sdd-lean return to sdd-orchestrate-flow, passing on --mode and --log-decisions when supplied, and the flow records the triage as a decision in workflow.md; for sdd-lean, also record the merge of HIL 1 and HIL 2 as a change to the stops. spot follows the branch below. Output: next step started on the decided path.

Spot branch

  1. Implement. Make the smallest coherent change with a regression test that fails before and passes after, or a behavior test when there is no defect. Validate with the repository's validation skills. Output: change applied and test result recorded before and after.
  2. Tripwire. Stop and return to step 2 with the new evidence when the diff exceeds the file limit decided at HIL 0 (default: three production files), touches S1, S2, S3, or S8, or a product decision appears. The change stays in the worktree as evidence; discarding it is a human decision. Output: diff confirmed within the limit, or triage reopened with the new evidence.
  3. Review. Review the diff with the repository's review skill or the host's native review. Output: review findings fixed or taken to the human.
  4. Deliver. Commit only when asked, through the commit skill. Output: delivery in the requested state, with the regression test and review as evidence.

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.