Delivery

Dely accepts a request that may still be vague, brings it to an approved design contract, then automates sequential implementation, independent review, and pull-request preparation. It is a thin control protocol, not an orchestrator, SDLC framework, or second source of Git state.

Read AGENTS.md in the repository for this project's gate commands, artifact paths, and default branch. Per-phase harness/model/effort pins come from .dely/, which dely pins resolves. This skill never names them.

Two human gates

1. Approve the design contract before candidate mutation.
2. Merge or publish after Dely has prepared the reviewed pull request.

Dely pauses outside those gates only for a scope or architecture change, a destructive action, new authority, replan, or an unavailable required runtime.

The control session

The current interactive session is Control. It owns the approval invariant, task boundaries, exception handling, dispatch supervision, and release. It does not implement or review the candidate, and does not prescribe question count, order, format, or skill-selection precedence. Native Plan Mode and active design skills govern their own surfaces under the harness's precedence; neither bypasses approval. Before mutation, Control obtains explicit human approval of a design contract. One approval covers that scope; a material change needs a new one. Control surfaces uncertainty that could materially change intent, acceptance, authority, public contract, architecture, or consequential risk. Material assumptions must be explicit.

Shape

Use the smallest contract that safely holds the change. Risk may promote an otherwise small change; diff size never demotes data-loss, security, permission, or public-compatibility risk.

Shape	Use	Artifact	Review
Spike	Investigation only; no candidate is delivered	Approved probe and recommendation; no delivery run	none
Bounded	Small change, clear behaviour and ownership	Approved in-chat design and short execution envelope	one whole-change review
Architectural	Multiple behaviours, public-contract change, architecture decision, or promoted risk	Approved decision record, task plan, execution envelope	task review per task, then one integration review

An approved design contract states intent and success criteria; scope and authority; affected public contract or architecture; consequential risks and material assumptions; and a plausible counterexample that distinguishes correct behaviour from a present-but-wrong implementation. If no executable instrument can discriminate, the contract names the manual inspection and its limit.

Architectural work uses templates/decision-record.md (durable) and templates/plan.md (transient, deleted in the release commit, before the release-binding review). Commit both before implementation begins — that commit is the review baseline.

Acceptance

One table: each requirement, the instrument that proves it, the plausible wrong implementation that instrument rejects, and where that rejection was observed.

An acceptance row is invalid until its instrument discriminates. A row that passes both before and after the change proves nothing and will be found at review. Baseline-red is not enough: red only because the feature is absent says nothing about catching an implementation that is present, runs, returns a pass, and is wrong. Each row names such a wrong implementation. "The feature is absent" does not satisfy Counterexample. Where none exists, the row says so and a human reads the diff.

Record what the instruments cannot observe. Prefer the simplest instrument that proves the contract. Any claim about extent — a scope, a count, a set of call sites — names the command that produced it, reports its output, and enumerates rather than samples. Naming the command is not the measurement; the command must have been run.

Execution envelope

Before mutation, Control resolves deployment preferences against the live harness surface, starts Orca and verifies its required capabilities, records the dirty baseline and exact-path ownership, and creates a feature branch when starting on the default branch. The envelope freezes owned scope and paths; protected pre-existing dirty paths; acceptance criteria, counterexample, and focused instruments; branch, base, remote, and pull-request target; resolved harness, model, and effort for dispatched roles, from dely pins --repo <path> with each phase's source (local, team or control); and authority to branch, commit owned paths, run gates, push, and open or update a pull request. It never authorises merge, force-push, stash, reset, cleanup, or an edit outside owned scope. Dely stages and commits only contract-owned paths. It never stashes, resets, cleans, or silently absorbs the user's existing changes. If a path carries protected baseline changes and Dely must also modify it, Control pauses rather than combining ownership.

Orca and the helper

Orca is the required execution plane. It launches and supervises fresh native harness TUIs with the resolved harness, model, and effort. Orchestration is a required Orca capability. dely:delivery starts Orca, then Control loads orca skills get orchestration and follows its supervised loop. It stops only when the CLI is missing, the runtime cannot start, or a required capability is absent — there is no direct dispatch and no headless fallback of any kind. The launcher is scripts/dely relative to this skill. Control learns the helper's interface by running scripts/dely with no arguments, which prints its identity and usage, and does not read scripts/dely.js. Control's wake mode is that harness's controlWake in ../../harnesses.json. The preflight step runs in setup and again after a NO_ACK; a delivery does not preflight before its first dispatch. Run dely preflight and dely dispatch with no tool timeout, or one of at least 300 s.

Write the prompt to a file in the Run's folder, <repo>/.dely/local/runs/<run>/, where <run> is the Orca Run id; dely dispatch refuses a --spec-file anywhere else. It writes .dely/local/.gitignore containing * when that file is absent and leaves an existing one as it is, so the Run's files stay out of git status only while that file ignores them. Never inline it in a shell argument: prompts carry backticks, quotes and newlines, and a shell argument mangles them. A path outside the workspace can trigger a second permission surface some harnesses still prompt for even when tool approval is skipped. The handoff is a file in the same folder; its path travels as payload.reportPath and the message body stays short. Routing, review and remediation read earlier prompts and handoffs, so nothing in the folder is deleted when a worker returns; it is kept until the run closes. A Spike that dispatches workers opens its own Run, dispatches through dely dispatch the same way, and closes when it reports its recommendation. --spec and --body are shell arguments, which this skill already forbids for prompts.

The dispatch prompt carries the task, its scope, and the evidence required. It does not define role dispositions or the conditions for reaching one — those belong to this skill, and a prompt that restates them narrows or contradicts them. Where the design contract states an acceptance row, the prompt carries that row as written. Every dispatch goes through dely dispatch. Control does not compose a worker launch or call worker-start by hand. The helper resolves the pins from .dely/ as dely:setup describes and appends the acknowledgement instruction and a sentence that the Orca preamble and the spec file are everything the worker needs and that it should read no other skill. The worker-start receipt records launch.requested and launch.effective; it does not establish that the worker can serve the request or that it cannot. Orca applies the execution plane's configured permission default and does not add a sandbox the project did not pin.

Name the model and effort on every dispatch. The helper passes --model/--effort when ../../harnesses.json says that harness takes them, and omits a flag whose value is default. A harness whose entry has effortInModel and modelFlag receives a pinned Effort joined to the Model by that string in --model (selector:level), never --effort; Effort default passes the Model alone, and a pinned Effort with Model default is refused. A harness whose entry names modelPin: spec-unchecked receives its pin as a spec line instead of flags, and the helper never verifies the reported model; a pin such a harness cannot resolve stops the worker before its first request and surfaces as NO_ACK. A spec line must not start dely-pin:. On a harness that takes neither flags nor a spec pin, a Model written there is silently not applied: write default and set the model in Orca's agent default arguments. A worker left on a harness default is an unpinned environment: it lives in the harness's own config, it changes without announcing itself, and the dispatch that relies on it looks identical to one that pinned the same value deliberately.

Never act on an Orca nudge. After DISPATCHED, wait by the wake mode of this Control's own harness — not the harness of the worker being waited on. --control is this session's harness id, the same id whatever worker is in flight. background runs dely wait --run <run> --control <self> as a background command and ends the turn; waker runs dely wait-bg --run <run> --control <self> as its last command, then ends the turn (a waker Control never runs dely wait); unsupported cannot be Control. The helper reads the harness of the Orca terminal it runs in and refuses a waker even when --control names another one, so REFUSED … (called with --control …) means use wait-bg.

Result handling. SETTLED: the helper has already released each worker whose accepted worker_done is in the batch and lists Orca's answer under release; Control does not release those itself and never closes a worker terminal by hand (terminal close). retained with user_takeover means a human typed in that tab; Orca keeps it and the human closes it. Process the batch, do the guide's completion accounting, and pass the settled deliveryId as --ack to the next dely wait or wait-bg, or run orca orchestration check --ack when no wait follows. When wait (or the wait-bg output) prints ERROR for --ack, run orca orchestration check --ack <settled id> and wait again without --ack; a waker Control waits again with wait-bg. ATTENTION has two routes, and the difference is whether the plane can still see the worker. With nextAction.kind other than none, run the argv Orca printed and skip that id next time. With nextAction: none and attention.requiresAction, the plane has lost sight of the worker rather than asked for something: read it with worker-read and worker-show, and if the process is gone, worker-stop, then worker-abandon when the stop reports stop_unknown, then worker-release, then one fresh dely dispatch with the same prompt file. A second time on the same input goes to the human. An absent nextAction is absent, not none with attention — that row is not ATTENTION and the wait continues. STALLED: read the output, then wait again or recover. NO_ACK: run setup's dely preflight in the same Run. If every pin passes, one fresh dely dispatch with the same prompt file; never retry into the same terminal, and never reuse a settled terminal; a second failure on the same input goes to the human. Any PREFLIGHT … FAIL: do not dispatch to any pin — a failed pin's cause is already known and another dispatch only repeats it. Stop and relay the printed reason: the harness, the path, and that the human opens that harness there once to answer its own dialog; Dely never answers it. The failed worker is already stopped and released. When the human says it is done, rerun dely preflight in the same Run and continue from there. FAILED: one fresh dely dispatch with the same prompt file and the same retry limits. DEADLINE: a checkpoint — check worker-list and the last output; if the worker is progressing, wait again; a second DEADLINE with no progress goes to the human. ERROR: go to the human. The worker reports once with worker_done and an --outcome; completion comes from that worker_done — do not infer it from reading the worker's terminal. Each delivery opens its own Run with an objective (orca orchestration run-create --objective) rather than reusing another's, so a stale report cannot settle a new wait.

Investigation is one read-only diagnostic dispatch inheriting the implement pin; it may reproduce, inspect, and report, but does not edit, commit, launch workers, or expand scope.

Stop and ask the human when: a result maps to no route or more than one; the worker failed rather than returned a stop status — a non-zero exit with no result, an exhausted quota, an authentication error — which is not BLOCKED and must not be treated as one; Orca is unavailable or a required capability is absent; an action needs authority policy reserves to the human; or the same worker fails twice on the same input. Say what you know, what you tried, and what the options are. Do not pick one.

Implementation

Control creates a separate task only when that unit has its own test cycle and a reviewer could accept it while rejecting its neighbor. Same-shaped mechanical changes are batched. Tightly coupled work stays one task and one implementer. Each independent task gets a fresh implementer TUI.

An implementer reads the decision record, the plan, and the baseline — not the design session's transcript. It owns only its task, runs a focused acceptance instrument, and creates one task-scoped commit. For behaviour with a deterministic executable test it uses TDD; the portable invariant is smaller: observe a discriminating failure for the intended reason before changing behaviour. A shell probe, parser fixture, or diff inspection may be the correct instrument for configuration, documentation, generated files, or environment-bound integration.

The counterexample named in each acceptance row is observed red and cited. That observation is not the behaviour's own absence: one is the feature absent, the other is an implementation that is present, runs, returns a pass, and is wrong.

Implement the whole task before handing back. Stop and return BLOCKED or NEEDS_REPLAN instead of a partial solution when the record contradicts the code, the contract is ambiguous, work outside the task becomes necessary, an existing test disproves an assumption, or the task no longer fits one session. A task needing continuation is a decomposition failure.

Handoff

Status: DONE | BLOCKED | NEEDS_REPLAN
Harness:            name, model, effort, sandbox
Session:            the dispatch id
Baseline:
Changed paths:
Contract coverage:
Verification:
Deviations from plan:
Residue:
Git state:
END OF HANDOFF

END OF HANDOFF is the last line and load-bearing: the only thing that distinguishes a handoff from one cut off mid-write. Under Residue, a claim of nothing left needs evidence: name the check that returned empty. Under Verification, cite the dispatch-bound command, output, and outcome that Orca recovers for that task — the transcript or terminal it selects, and any cursor mechanics, are Orca's concern, not this skill's. Do not transcribe output by hand. Where Orca cannot recover a dispatch item, treat the worker's own account as the thing under check rather than as the check, and say so.

Review

Review independence is role independence: a fresh session that did not implement and does not edit the candidate. It gets the decision record (or Bounded design), the baseline, and the diff. The phase adds no sandbox by default; a project may pin one for a concrete risk in AGENTS.md. Review depth is adaptive: Bounded work gets one independent whole-change review. Each Architectural task gets an independent task review. After all tasks are accepted, a different fresh reviewer performs one integration review of task interactions, complete-contract coverage, deferred findings, candidate identity, and release readiness. Only that final review is release-binding for Architectural work; Bounded work has no earlier task review and no duplicate integration review. No worker runs while a review of the same working tree runs. The working tree and its gate surface are shared mutable state, and a review reproduces gates in that tree, so a concurrent edit makes another task's work look like this one's result.

Reproduce, do not accept. Run the gates yourself. A claim you did not reproduce is not evidence. The reviewer observes the counterexample discriminate for itself — an implementation that is present, runs, returns a pass, and is wrong. An instrument red only because the behaviour was absent is not that observation.

Classify findings: Blocking — contract failure, regression, data or security risk. Important — missing required behaviour, test, or reconciliation. Minor — useful, does not block. Out of scope — recorded, not absorbed. Return exactly one role disposition: ACCEPT, CHANGES_REQUESTED, or BLOCKED. State what the review did not verify — what it did not reproduce or read. A contradiction you cannot resolve is CHANGES_REQUESTED. Do not open remediation over wording when deterministic checks already prove the contract.

Remediation

One pass is one remediation pass per finding, not per review. The original parties are defined by pin and briefing, not by session: the helper releases every worker at settlement, so remediation and re-review are fresh dispatches briefed from the Run's folder.

For an in-contract CHANGES_REQUESTED finding, remediation is a fresh dispatch on the implement pin whose prompt names the original task prompt, the implement handoff, and the finding as the reviewer wrote it. That worker is the original implementer for every rule here: it verifies the finding, fixes the root cause, reruns the affected instruments and closure gates, and writes a separate remediation commit. Control owns the plan and the decision record for the whole run, including remediating findings inside them; amending them is not implementing the candidate.

The scoped re-review is a fresh dispatch on the review pin whose prompt names the original review prompt, the raised finding's handoff unedited, the fix-only diff, and the full candidate diff for context; a re-review not briefed with that handoff is not the scoped re-review. It checks the finding's reproduction, the fix, and any defect the fix introduced, and scope-checks a Control amendment the same way. If it does not accept, Control routes to REPLAN_OR_SPLIT — it does not start another repair loop. BLOCKED preserves the candidate and escalates the unresolved dependency or authority question to Control separately from a failed worker process; it is not remediated. The Architectural integration review is always a fresh reviewer.

Release

Control performs release with native Git and forge tools; release dispatches no LLM worker and makes no post-review candidate edit.

1. Complete implementation and, for Architectural work, its task reviews.
2. Reconcile owning documentation, delete the plan, and commit the complete candidate.
3. Run the focused instruments and project closure gates on exact HEAD.
4. Push the feature branch and create or update a draft pull request.
5. Run the applicable final review while remote CI runs on that same HEAD.
6. Require both that review's ACCEPT and required checks green.
7. Mark the pull request ready and report it for human merge.

Any candidate mutation after the applicable final review invalidates that verdict; Control reruns the affected gates and review on the new exact HEAD. Affected gates are those that can observe the change class; a project may name that subset. Dely never merges, force-pushes, or publishes outside the approved target and authority. If project policy cannot publish work in progress, Dely delays the push and pull request until the applicable review accepts.

Maintenance logging is machine-local and opt-in at ~/.dely/log.jsonl. It stays opt-in on the presence of ~/.dely/ and is never created by Dely. At close, after release or when the run stops early, Control deletes the Run's folder <repo>/.dely/local/runs/<run>/, then runs dely log --run <run> --repo <path> --json '<object>' rather than assembling a line by hand. The log adds residue, the files still in that folder ([] when none, null when no repository is known or when part of the folder could not be read, which also adds residueError, the error code and the path it failed on); it deletes nothing. It also releases every worker Orca still lists as reclaimable for the Run and records each answer under release (releaseError when the listing failed). dely with no arguments prints which copy is running, and its usage.

Failure and recovery

Recovery uses Orca records, Git, CI, and pull-request state — never inferred from an ambiguous, missing, or merely transport-level outcome.

Failure	Disposition
In-contract implementation defect	Original implementer remediates
Scoped remediation re-review does not accept	REPLAN_OR_SPLIT
Scope or architecture must change	Return to the design gate
New authority or destructive action is required	Ask the human
Orca or a required capability is unavailable	Stop; no headless fallback
any PREFLIGHT … FAIL	Dispatch to no pin; relay harness, path and dialog to the human; rerun preflight in the same Run when told
Harness fails or evidence is insufficient	Preserve the candidate, report the native outcome and role disposition
Idempotent release step is interrupted	Verify Git and pull-request state, then resume