Skip to content

Latest commit

 

History

History
106 lines (86 loc) · 6.72 KB

EXECUTION.md

File metadata and controls

106 lines (86 loc) · 6.72 KB

Release execution

A person opens the public website, understands what they can use, and chooses Home on the same server or installation on their own hardware. Installation instructions identify the published download and development build separately. The server runs a reviewed candidate assembled from the active source work. Its deployment receipt identifies the source and served artifacts. A public download version has a separate verification record.

Record ownership

Record Owns
GitHub issues Scope, acceptance, status, dependencies, owner and proof
Journey audit workbook Historical journey and finding snapshot; test input only
Private current checkpoint and receipts Worktree/target coordinates, resource owners, exact commands and raw proof
Notion overview Accepted milestone summaries and links to reviewed source

The coordinator serializes status updates. Each implementer owns a bounded set of source files and the complete repair/test cycle. Another agent reviews the actual diff and evidence. Human acceptance retains its human author.

Use the development-loop procedure for dependent work. Keep one coordinating mission goal and the existing journey IDs. The current checkpoint is development-loop-current.md in the Git common directory. Resolve that directory with git rev-parse --path-format=absolute --git-common-dir before resuming. Read the latest user correction, then verify the named checkout and target.

First pilot

The first work is a useful storefront and a human Home entry point. This gives the execution procedure a real product result to verify while the remaining release work keeps its full acceptance.

Milestone Observable acceptance
C1-site The landing page explains the product, opens /home/, works at desktop and mobile widths, and supports keyboard use. The visitor page focuses on Home and useful actions. Source, hosted Home and published installer identities remain distinct in operator records. Installer promotion requires accepted release and platform evidence.
C1-home /home resolves to /home/; index and relative assets use the existing Home capsule handler and response policy. Old root bookmarks reach the new entry. Sign-out returns there. Capsule launch identity, API paths and Runtime authority remain coherent.
C1-proof Source checks, actual rendered success/failure states and an independent review pass at the named revision. The process resumes at the same next action and detects stale proof, owner conflicts and repeated failures in bounded scenario tests.
C1-seed An approved candidate is installed on the seed with preserved identity and user state. Site and Home artifacts match its receipt. The public root and /home/ work; sign-in, sign-out and an ordinary app launch have target evidence.

C1-site can be reviewed as its own source slice. Public deployment follows the target's acceptance and approval gate. The full signed three-platform installer and update journey remain part of the release, even when this first pilot passes.

Verification recipes

Start with the smallest relevant failure: a metadata response, one Home route, or one installed interaction. Then run the changed surface through its consumer.

git diff --check
node scripts/check-product-data.mjs
node scripts/website-truth-check.mjs
(cd elastos && cargo fmt --all -- --check)
cargo fmt --manifest-path capsules/chain-provider/Cargo.toml -- --check

The owner adds narrow route tests and the existing Home/browser smoke checks that cover the change. The owning issue records expected route changes and results. Preserve existing Pass expectations and each failed run's scope. Source checks qualify source; installed and public claims require matching artifacts and real target behavior.

For the website, inspect desktop and mobile layout, keyboard tabs/actions, missing metadata and inconsistent head/release data. Resolve local assets and the exact action URLs. A visual prototype is evidence about the page only.

For Home, test the index, relative modules, manifest, old bookmark redirect, sign-out, authenticated API/event access and ordinary app launch. Before a target test, bind the source, built binary, installed binary and capsule tree. Record a restart and the observed result through the public route.

Monitoring decisions

The existing heartbeat reads the current checkpoint and changed evidence. It observes the active coordinator and its explicit owners. Old paused tasks keep their pause until an assignment transfers their work. Before a stop correction, verify the original user message's date, task and scope against the current mission. A historical stop in a reused monitor task does not cancel a later authorized mission. Report uncertainty when the source instruction cannot be established. The monitor writes observations and pending corrections to its private shared state in a writable operator folder. The coordinator reads that record at each changed milestone and every 30 minutes, then records adoption in its checkpoint. This keeps observation and implementation ownership separate without requiring the monitor to send task messages or change repository metadata.

An evidence checkpoint is due every 30 minutes during long work. Report passed, failed and pending milestones, the newest useful result and the next experiment. Measure time to useful proof, repeated experiments, reopened acceptance and reviewer corrections. Commit count and status edits measure activity only.

Observation Next action
Source/artifact changes after proof Mark affected proof stale, preserve history, recheck affected consumers.
Same failure and unchanged experiment twice Diagnose the repeated boundary and choose a different small experiment. Check whether a previous correction was adopted before sending it again.
Long build with current output Retain its owner; report progress and its next evidence checkpoint.
No useful result and no live work at the checkpoint Inspect the owner and prerequisites; record a concrete next experiment or missing resource.
Two writers or target owners overlap Preserve the active owner; queue the dependent operation and continue independent work.
A new feature or weaker acceptance enters the diff Restore the agreed scope or seek the user's decision with the concrete change.
Native task retrieval fails Record the observation limit, read the short checkpoint and named receipts, and avoid treating missing telemetry as failure.

The monitor reports meaningful accepted progress, material drift/failure and required user input. It remains quiet during unchanged routine work. Local scheduled monitoring requires the app and host to be running.