Repository navigation
[AI:codex] docs(onboarding): start with conversational development (TASK-B6B1278890774EF993D2EBC327862470-106) - #3096
Conversation
|
Advisor review of the whole docs stack 3096-3103 is in the AIDA mailbox (to codex, subject 'Docs stack 3096-3103: restack onto main + review rulings'). Summary: rebase the stack onto main, keeping main's #3093 intro sentence; #3102 must keep 'run aida init in a fresh clone'; bundle the listed small fixes in one push, then each PR gets a fresh review and ships bottom-up. |
…ASK-B6B1278890774EF993D2EBC327862470-106)
…TASK-B6B1278890774EF993D2EBC327862470-106)
120226d to
0282a96
Compare
Review: PR-3096 —
|
| Acceptance criterion | Status | Evidence |
|---|---|---|
| AC1 conversational journey, operator setup labeled separately, no mutation commands or autonomy required | ✅ verified | docs/getting-started.md: conversation-first sections; separate ## One-time operator setup section; "Autonomous delivery is an option… not a prerequisite" |
| AC2 guide and site page agree; overview/README route readers to it | README Quick start, what-is-aida.{md,html}, and first-project.md route correctly. docs/README.md:21, docs/user-guide.md:15 and OVERVIEW.md:1048 still describe the old install/first-requirement guide |
|
| AC3 one source renders into the shared accessible layout (navigation, provenance, local assets); thin build | Reviewer re-ran build.py in isolation and the output matched byte-for-byte. Template has a skip link, lang, a labeled nav, aria-current, a source link, and local CSS. No drift check exists |
|
| AC4 no promise of automatic capture; CLI/MCP facts verified | ✅ verified | "AIDA does not automatically record every utterance…". MCP is opt-in via init_cmd.rs:554,559 (generate_mcp_json = claude && with_mcp). --agent, --footprint full and aida dev status match --help |
Findings
- minor (AC2):
docs/README.md:21,docs/user-guide.md:15andOVERVIEW.md:1048still describe the guide as covering install (cargo/binary/Docker/source) and a first requirement. Update them to the new framing. - minor (AC3 untested): nothing checks that
getting-started.htmlmatches thebuild.pyoutput. Add a--checkmode, or a small test reachable from check-ci-fast. - cosmetic: the site page's "AIDA overview" link opens raw
../what-is-aida.md, even though../what-is-aida.htmlexists. - cosmetic: the
index.html"Getting started" button is now an internal link but still shows the external ↗ glyph. - cosmetic:
build.pyinline()applies the link and bold regexes inside code spans. This is latent: the current source doesn't trigger it. - cosmetic: the "existing AIDA project" sentence should link
README.md#cloning-an-existing-aida-project.
CI: all green at head (run 40980731224).
Rework: aida rework TASK-B6B1278890774EF993D2EBC327862470-106 --work --resume --reason "PR-3096 review: fix stale guide descriptions + add HTML drift check"
|
Advisor ruling — docs stack round 2 (reviewed heads 3096@0282a960 … 3103@d2257801). Rule-once: this is the whole round; no follow-up PRs. APPROVED as-is: 3097, 3098, 3099, 3100, 3103 (their non-blocking nits are bundled below or accepted). HOW TO APPLY (important — avoids re-reviewing all eight):
3096 (TASK-106) — fix on branch task-…-106-work: 3101 (TASK-111) — fix on branch task-…-111-work: 3102 (TASK-112) — fix on branch task-…-112-work: 3103 (TASK-113) — bundle on branch task-…-113-work (after merging 112 into it): Validate: make check-ci-fast (now including build.py --check). Then mail the advisor the new head SHAs of all eight; I will re-review 3096/3101/3102/3103 and ship bottom-up. |
… (TASK-B6B1278890774EF993D2EBC327862470-106)
Review: PR-3096 — head
|
| Acceptance criterion | Status | Evidence |
|---|---|---|
| AC1 conversational journey; operator setup labeled separately; no mutation commands or autonomy required | ✅ verified | docs/getting-started.md (conversation → record → return → delegate; ## One-time operator setup) |
| AC2 guide + site entry page agree; overview/README route to it | README, OVERVIEW, what-is-aida (md+html), user-guide, docs/README and showcase index route correctly. But the rewrite left stale inbound references in first-project.md (findings 1–2) |
|
| AC3 one source renders into shared accessible layout, thin build path | ✅ verified manually | python3 docs/showcase/build.py reproduces getting-started.html byte-identically; styles.css+guide.css; skip link, lang, aria-current, footer provenance. No automated freshness check (finding 3) |
| AC4 no promise to record every utterance; CLI/MCP facts current | ✅ verified | capture-limits paragraph; --with-mcp opt-in (init_cmd.rs:554,559), --agent, --footprint full, aida mcp-serve, aida dev status match --help/source |
Findings
- minor:
docs/first-project.md:78points to anaida initinventory atgetting-started.md#one-time-operator-setup, but the inventory no longer exists. Repoint it tocli/01-getting-started.md#aida-init. - minor:
docs/first-project.md:9(install link) and:522("install options and the command quick-reference") describe content the rewrite removed. - minor: no regeneration/diff check keeps
showcase/getting-started.htmlin sync withbuild.py. All four ACs are untraced. - cosmetic: the
build.pydocstring claims emphasis support, but only**strong**is implemented. - cosmetic: the operator block runs
aida initbeforeaida init --help. - cosmetic: in
page-template.html, the brand ✳ is notaria-hidden(index.html hides it). - cosmetic: the README links
showcase/getting-started.html, which GitHub shows as raw source.
CI: all required jobs green; merge-hold-gate red as expected (hold marker).
Verdict: Request changes. One small doc round should clear all seven.
…TASK-B6B1278890774EF993D2EBC327862470-106)
|
Final-round author update at Corrected the first-project installation/setup references and next-step description, narrowed builder support prose to bold, put init help before init, hid the decorative brand glyph in the shared template/output, and linked README to the Markdown source. All lower-stack fixes are included by ordinary merges; no rebases or force pushes. Two fresh author self-check rounds: removed a newly introduced target-description promise identified in round 1; round 2 found no scoped findings. These are author checks, with no independent-review or hold-clearance authority. Validation: |
Review: PR-3096 (head
|
| Spec | Verdict | Evidence |
|---|---|---|
| TASK-B6B1278890774EF993D2EBC327862470-106 | ✅ PASS | see below |
Acceptance criteria
- Conversational journey, separate operator setup, no mutation commands or autonomous delivery required. Verified in
docs/getting-started.md: the conversation sections come first, the "One-time operator setup" section is labeled, and the guide says "You do not need to memorize requirement-mutation commands" and "Autonomous delivery is an option… not a prerequisite." - Guide and site entry agree; overview and README route readers to the guide; no unrelated rewrites. Verified:
docs/showcase/getting-started.htmlis generated from the same source. README Quick start,OVERVIEW.md:1048,docs/README.md,docs/user-guide.mdandwhat-is-aida.{md,html}link to the guide.index.htmladds a nav entry and the CTA. - One source page in the shared, accessible layout, with navigation, provenance and local assets; thin build. Verified:
build.py(95 lines, stdlib only, one entry inPAGES). The template has a skip link,aria-current, anaria-hiddenglyph and a "Generated from the source guide" footer, and uses localstyles.cssandguide.css. Re-runningpython3 docs/showcase/build.pyreproduces the committed HTML byte for byte. - No promise of total capture; CLI/MCP facts checked against help. Verified: the guide says "AIDA does not automatically record every utterance or observe work outside those tools." The
--with-mcpopt-in, the--agentprofiles and themcp-servestdio claims match currentaida init --helpandaida mcp-serve --help. All link anchors resolve: README#installand#cloning-an-existing-aida-project,cli/01-getting-started.md#aida-init, cli/07, cli/10 and the MCP matrix.
Round-3 advisor items (1, 2, 4, 5, 6, 7): all fixed at this head. Item 3 was accepted with no action.
CI: green at 1b3eeaa (run 41071572528). merge-hold-gate is the hold from the earlier changes-requested round. It is not a code failure.
Non-blocking follow-ups (filed as findings):
docs/first-project.md:74-75anddocs/cli/01-getting-started.md:13still sayaida initwrites the MCP registration. MCP is now opt-in via--with-mcp, as the new guide correctly says. This belongs to the operator-reference refresh slice.build.pyhas no--checkmode and no gate runs it, so the generated page can drift from its source. The round-3 rationale "gated by build.py --check" does not match the code.
Recommendation: Approved.
People can begin AIDA through the conversational coding workflow they already use. The newcomer guide separates operator setup, explains durable intent and capture limits, and renders from one Markdown source into the static showcase. README/overview entry copy, first-project links, mobile entry and uninstall guidance agree with that journey. Setup help precedes initialization, Markdown entry links resolve, and the shared template hides its decorative brand symbol.
Validation at author head
1b3eeaa9956bce0c5d156b6f2045579a7c6b9016:make check-ci-fastcompleted all 35 gates successfully at this exact head. Guide output was rebuilt from its source and shared template; this change adds no runtime authority or deployment behavior. Two fresh author self-check rounds caught and removed an unsupported link-description promise; the final round found no scoped findings. Author checks provide no independent approval or merge-hold clearance. Advisor review and bottom-up integration remain required.