saf-implement git:20260826.655710eA

Immutable. This exact content is served forever at /api/v1/blob/0c6e49fb2d1dc50f.

---
name: saf-implement
description: Implement exactly one validated SDD task as the smallest tested, merge-ready increment. Use for a single task reference or explicit task implementation request; not for planning a feature or coordinating several tasks.
compatibility: Requires Git and a compatible Agent Skills host.
---

# Implement one SDD task

## When to use

Use for one unambiguous task that is ready to implement or resume. Read [the TLC baseline](../sdd-agentic-flow-shared/references/tlc-baseline.md), [the TDD baseline](../sdd-agentic-flow-shared/references/tdd-baseline.md), [task slicing](../sdd-agentic-flow-shared/references/task-slicing.md), [feature profiles](../sdd-agentic-flow-shared/references/feature-profiles.md), [engineering principles](../sdd-agentic-flow-shared/references/engineering-principles.md), [spec lifecycle](../sdd-agentic-flow-shared/references/spec-lifecycle.md), and [safety rules](../sdd-agentic-flow-shared/references/workflow-safety.md) before acting.

## When not to use

Do not use for specification authoring, several tasks, a feature-wide validation, PR review, or a task whose identity, scope, or dependencies are ambiguous. Reject a discovery-only workspace: it is not a spec package and must first pass through `saf-create-spec`.

## Inputs

- A single canonical task reference or explicit feature and task identifiers.
- Repository SDD artifacts, relevant code, and optional `.sdd-agentic-flow/config.yml` overrides.
- Optional task prompt or prior handoff, treated as supporting evidence only.

## Workflow

1. Read `.sdd-agentic-flow/config.yml` when present; otherwise use canonical effective defaults.
2. Read `.sdd-agentic-flow/context/project-context.md` and `.sdd-agentic-flow/context/domain-glossary.md` when they exist. Read `workflow.feature_profile` from `.sdd-agentic-flow/config.yml` and apply feature-profile guidance for evidence rigor. Resolve exactly one package, then exactly one task from the configured SDD source. Load the [Task Context Package](../sdd-agentic-flow-shared/references/task-context-package.md) minimum context. Resolve bounded goal, completion criteria, constraints and applicable [decision gates](../sdd-agentic-flow-shared/references/decision-gates.md) before mutation per [bounded-execution.md](../sdd-agentic-flow-shared/references/bounded-execution.md). Load this skill's existing Inputs/Workflow list only; related slugs only if named or requested (one hop). Confirm its acceptance criteria, requirement anchors, dependencies, allowed scope, and current implementation state.
3. Inspect callers and existing patterns before editing. Stop if the work requires a sibling task, unsafe environment, or unresolved authority conflict. Specifications are **living** control artifacts: if you find drift, first determine whether repository evidence supports an intent-preserving reconciliation. In autonomous mode, route that reconciliation to `saf-create-spec`, refresh affected prompts, and resume without asking when acceptance semantics and delegated scope remain unchanged. Do not silently implement a different requirement, silently rewrite the spec from implementation preference, or proceed when the change expands authority. A contract-change proposal is non-authoritative and never authorizes its own semantics.
4. Apply `../sdd-agentic-flow-shared/references/engineering-principles.md` before editing. Search existing patterns, prefer modifying an existing file, and keep the complexity budget. In autonomous mode, choose an evidence-backed technical design that preserves the effective contract; do not add a competing architecture, new dependency, or new convention when the decision would expand delegated authority.
5. Identify the required behavior from the spec, the contractual seam (field label: `Public seam`; prefer public/observable when practical), the sensor, and the oracle/acceptance condition from spec, repo contracts, or configured gates — never solely from the implementation. Stop when the seam is unclear. Stay inside the fix boundary; do not expand into **unchanged behavior**. For bugfix or refactor intent, record regression sensors. Do not complete an **investigation** as a fix (`Status: pass` on findings is forbidden).
6. Use one vertical slice at a time: name the behavior, place a sensor at the contractual seam, implement the smallest change, and record executed **current** evidence. Test-first is recommended when it sharpens the spec. Full RED → GREEN → REFACTOR is optional and is never harness proof. Do not fabricate RED. Do not weaken required behavioral coverage because the ritual is optional. Do not complete on self-assessment. Do not achieve green by suite weakening (deleting, skipping, or narrowing tests that encoded the AC). The oracle stays the spec-derived expected outcomes.
7. Apply `../sdd-agentic-flow-shared/references/evidence-standard.md`. Record commands, results, limitations, and untested risks. A passing sensor is evidence, not a correctness verdict. Do not claim done because the conversation feels finished.
8. Report TDD evidence, changed files, checks, remaining risks, and the next SDD step. Do not commit, push, open a remote PR, or update external trackers unless the user separately asks.

## Safety

Preserve unrelated and pre-existing changes. Keep credentials, personal data, and local paths out of output. Do not mutate production, remote services, tracker state, Git history, or repository configuration by default. Do not complete on self-assessment. Do not perform suite weakening to obtain green.

## Output

Return the resolved task, outcome (`implemented`, `partial`, `blocked`, or `no changes required`), concise evidence, validation results, and recommended next step. Include `Status`, `Next recommended skill`, and `Reason`. Terminal output is an **implementation candidate ready for verification**, never its own final verification verdict. In autonomous mode, continue through another bounded attempt when evidence shows semantic progress and the transition is admissible; return control only for an exceptional authority or safety boundary, exhaustion, or no progress. When work pauses before a terminal outcome — session end, an agent swap, or a blocker only a human can resolve — write or update `handoff.md` per `../sdd-agentic-flow-shared/references/handoff-standard.md`.

### Autonomy

Supports `manual`, `supervised`, and `autonomous` autonomy levels. In `autonomous` mode, tests,
configured lint, and semantic scope remain required before a normal check transition. A recoverable
failure authorizes an owning repair or retry when semantic progress remains; an authority or safety
boundary, exhaustion, or no-progress result returns control to the human. See
`../sdd-agentic-flow-shared/references/autonomy-guardrails.md`.