Repository navigation
fix(rhodibot): stop demanding A2ML; make CONTRIBUTING/CODE_OF_CONDUCT .adoc|.md toggleable - #678
Conversation
… .adoc|.md toggleable The estate has retired A2ML: repo facts are clauses of <repo>_chora.deed. rhodibot still demanded nine .a2ml files (0-AI-MANIFEST, STATE, META, ECOSYSTEM, AGENTIC, NEUROSYM, PLAYBOOK, ANCHOR, rsr-profile) and, in `canon`, CONTRIBUTING.md / CODE_OF_CONDUCT.md. - New src/deed.rs: read-only s-expression reader for the chora deed (top-level clauses, (rsr-profile ...), (bot-directives (bot :name rhodibot (config ...)))). deed_lint.js stays the conformance gate; rhodibot only reads structure. - New src/canon/overlay.rs: re-targets criteria 2.1.4, 2.1.5, 2.3.1, 3.1.2-3.1.8 and 3.2.2 as requirements (deed clause or descriptile .deed), suspends 3.1.9 and 3.2.1, and adds a chora-deed check. A legacy .a2ml is a read fallback only, reported as "migrate" and never named as the form to create. The vendored rsr-criteria-v2.a2ml is untouched; the divergence is recorded under [overlay] in canon/pin.toml and checked by a lockstep test. - The profile now comes from the deed's (rsr-profile ...) clause, with the legacy file read only when the deed has none. - CONTRIBUTING / CODE_OF_CONDUCT: either .adoc or .md satisfies, never both required. Default is .adoc; (config :community-doc-format md) in rhodibot's bot directive makes .md the named form. A bad value is an error. - `check` (rsr.rs), fleet rule ids/suggestions and the webhook checklist drop the .a2ml/.scm entries and require <repo>_chora.deed instead. - .claude/CLAUDE.md invariant 1 rewritten: deeds, not A2ML. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ
|
Caution Review failedThe pull request is closed. Walkthrough
Merge Risk:
|
…) clause - (rsr-profile ...) accepts :spec and :declares-against, is exactly once, and needs :capabilities or :preset. - 3.1.2 state: descriptile STATE.deed first, fallback chora (status ...) or (maturity ...); there is no (state ...) clause. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ
… doc toggle - (rsr-profile ...) known keys now include :policy and :omitted-rationale, as in the estate's real clauses; nested children are ignored. A test parses a verbatim copy of standards' own clause. - 2.1.3 / SECURITY follows the same .adoc|.md toggle as CONTRIBUTING and CODE_OF_CONDUCT, in both `canon` and `check`. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ
…g-file error
Criterion 9.1.4 ("Deployed site serves its .well-known/ over HTTPS") is a
deployment question that the canon reader turned into a file check for
`.well-known/`, flagging templates that serve it from www/. It is now listed in
overlay::MANUAL_ONLY (recorded in pin.toml) and reported as a `manual` finding
("manual check: not verified by rhodibot"): visible, not graded, never tripping
--fail-on. 1.2.4 (.tool-versions) is detect=manual but a real file question, so
the list is explicit rather than a rule. The fleet report-back driver maps
manual to info.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ
A manual finding had no path, so the fleet grid would grade it against every change. It now carries the first path its description names (`.well-known/` for 9.1.4), else the canon's template path. Test asserts every manual finding has a non-empty path. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ
|
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d17c9955e2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| /// community documents whose spelling (`.adoc` or `.md`) is toggleable. | ||
| fn community_stem(name: &str) -> Option<&str> { | ||
| name.strip_suffix(".adoc") | ||
| .filter(|stem| matches!(*stem, "CONTRIBUTING" | "CODE_OF_CONDUCT" | "SECURITY")) |
There was a problem hiding this comment.
Keep SECURITY outside the community-doc toggle
When a deed sets :community-doc-format md, this match also reorders the SECURITY candidates, so a missing security policy is reported as SECURITY.md and the fleet suggestion claims the directive prefers Markdown. The documented directive only controls CONTRIBUTING and CODE_OF_CONDUCT, while SECURITY.adoc remains the estate default; restrict this match to those two documents. The canon overlay has the same unintended behavior through doc("SECURITY").
Useful? React with 👍 / 👎.
| // A copy only under a retired location is already "move | ||
| // it"; the new form is still what to move it to. | ||
| deprecated @ GroupVerdict::Deprecated { .. } => deprecated, |
There was a problem hiding this comment.
Preserve the deed destination for retired legacy files
When the only legacy file is under a retired location such as .machine_readable/6a2/STATE.a2ml, returning the original Deprecated verdict discards the replacement alternatives. Finding::of consequently reports the legacy A2ML path as deprecated but provides no STATE.deed or chora-clause destination, unlike the normal legacy fallback, so users are told to move a retired format without being told to migrate it to the deed form. Preserve alternatives in the resulting verdict, for example by converting this case to an Elsewhere result that still records the retired copy.
Useful? React with 👍 / 👎.
| } => { | ||
| expected.extend(alternatives.iter().cloned()); | ||
| let mut line = format!("absent — expected {}", alternatives.join(" or ")); | ||
| expected.extend(alternatives.iter().map(|path| describe(path))); |
There was a problem hiding this comment.
Keep machine paths separate from clause descriptions
When a required deed clause such as (manifest ...) is absent, this stores human prose like the (manifest ...) clause of widgets_chora.deed in expected. I checked the rhodibot arm in scripts/fleet-report-back.sh, which uses .expected[0] as the finding's file, while docs/REPORT-BACK.adoc requires a repository-relative path for diff-scope matching. The resulting error therefore cannot match a change to widgets_chora.deed, so removing a required clause can leave the gating finding outside the PR delta; retain the actual deed path in expected and use describe only for display text.
Useful? React with 👍 / 👎.
| Some(')') => bail!("unbalanced parentheses: a closing parenthesis with no opener"), | ||
| Some('"') => self.string(), | ||
| Some(_) => Ok(Node::Atom(self.atom())), |
There was a problem hiding this comment.
Parse quoted DEED profile values
For a valid profile that uses a quoted DEED value such as (rsr-profile :capabilities '(rust)), node parses the quote as a separate atom and the following list as another node, so rsr_profile sees :capabilities followed by "'" and rejects it as non-list. Quoted symbols and lists are part of the DEED value grammar already implemented in bots/panicbot/src/deed.rs; such a repository makes the canon command abort instead of producing a report. Handle the quote prefix and return the underlying symbol or list node.
Useful? React with 👍 / 👎.
| let mut legacy_profile = None; | ||
| for candidate in LEGACY_PROFILE_PATHS { | ||
| if let Some(text) = client | ||
| .get_file_content_if_present(owner, repo, candidate) | ||
| .await? |
There was a problem hiding this comment.
Delay legacy-profile reads until the deed needs them
When a repository has a valid deed-level (rsr-profile ...) but retains an unreadable or API-failing legacy profile, this unconditional loop still fetches the legacy file and aborts the canon command even though CanonReport::build_with will never use it. The local path has the same behavior because local::read always calls profile(root)?. This makes migrated repositories depend on stale fallback data; inspect the deed first and read the legacy candidates only when its profile clause is absent.
Useful? React with 👍 / 👎.
| self.root | ||
| .members() | ||
| .iter() | ||
| .filter_map(Node::head) | ||
| .map(str::to_string) |
There was a problem hiding this comment.
Distinguish list-valued fields from child clauses
For a valid top-level list field such as :labels (manifest), the generic tree stores the field value as a List, and this iteration treats that list's first atom as a top-level clause. The overlay then creates a virtual #manifest path and incorrectly marks criterion 2.3.1 satisfied even though no (manifest ...) clause exists; the same collision can occur for the other migrated fact names. Preserve the DEED distinction between :key value fields and nested clauses instead of classifying every list member by its head.
Useful? React with 👍 / 👎.
| let directives = self.clause("bot-directives")?; | ||
| directives | ||
| .children("bot") | ||
| .find(|node| node.field(":name").and_then(Node::text) == Some(bot)) | ||
| .map(|node| BotDirective { node }) |
There was a problem hiding this comment.
Reject ambiguous duplicate rhodibot directives
When a deed contains two (bot :name rhodibot ...) entries with conflicting :community-doc-format values, this find silently selects the first one, making behavior depend on source order instead of reporting the ambiguous configuration. Multiple (bot-directives ...) or (config ...) clauses are similarly reduced to the first match. The deed reader already rejects duplicate rsr-profile clauses because silently choosing policy is unsafe; apply the same uniqueness validation before consuming rhodibot's directive.
Useful? React with 👍 / 👎.
) ## What Follow-up to #678. standards#1605 (merged 2026-11-15 10:39 UTC) defines the `(rsr-profile ...)` key set; this makes rhodibot's reader match it, as A7 asked. ## Changes (`bots/rhodibot/src/deed.rs`, tests only otherwise) - Key set is now exactly `:version :spec :declares-against :role :preset :capabilities :add :remove` (`RSR_PROFILE_KEYS`), matching `bun 1-formats/deed/tools/deed_lint.js --rsr-profile-keys` at standards 10:39 UTC. `(notes …)` is the one nested clause and is not scanned. - Dropped the keys #678 had added for other clauses: `:policy` (belongs to `(anchor …)` and `(egress …)`), `:omitted-rationale` (belongs inside `(notes …)`), and `:schema-version`. Any other top-level key still fails closed. - `:capabilities`, `:add` and `:remove` must be lists of **strings**; symbols are now refused, as the module requires. A7 confirmed the real template conforms (`deed_lint.js --rsr-profile` passes on rsr-template-repo main), so this should not change rhodibot's result on the template. I have not re-run rhodibot against the template myself; SysArch's driver run would confirm. ## Tests `cargo test` in `bots/rhodibot` passes (172 + 11 + 58 + 1); `cargo fmt` and `cargo clippy --all-targets` clean. New tests pin the key set, the refused keys, `:omitted-rationale` inside `(notes …)`, and string-only lists. The verbatim standards clause test still passes. ## Not done here Left for A7's follow-up PR to #1605 (does not change the key set): empty `:capabilities` with no `:preset` refused, `(notes …)` fields once each, profile only on `*_chora.deed`. Rhodibot still accepts an empty `:capabilities` until that is ruled. --- _Generated by [Claude Code](https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ)_ Co-authored-by: Claude <noreply@anthropic.com>



What
rhodibot reported 12 errors on
rsr-template-repo: nine demanded.a2mlfiles the estate has retired, two demandedCONTRIBUTING.md/CODE_OF_CONDUCT.md, one demanded.tool-versions. This removes the nine A2ML demands and the two.mddemands.How
src/deed.rs(new): a read-only s-expression reader for<repo>_chora.deed: top-level clauses,(rsr-profile ...), and rhodibot's(bot-directives (bot :name rhodibot (config ...)))entry. Conformance staysdeed_lint.js's job (standards CI); rhodibot only reads structure.src/canon/overlay.rs(new): re-targets 12 criteria as requirements (2.1.3, 2.1.4, 2.1.5, 2.3.1, 3.1.2–3.1.8, 3.2.2), suspends 2 (3.1.9 CLADE, deed form not landed; 3.2.1), and adds a "chora deed exists" check. Deed forms:(manifest …), descriptileSTATE/META/ECOSYSTEM/AGENTIC.deedfirst with the chora clause as fallback (state:(status …)/(maturity …)),(neurosym …),(playbook …),(anchor …),(rsr-profile …). A legacy.a2mlis only a read fallback, reported as "migrate" (never "missing") and never named as the form to create.rsr-criteria-v2.a2mlkeeps its digest. The divergence is recorded under[overlay]incanon/pin.tomland a lockstep test fails if the record and the module disagree. Converting the canon to a deed is A7's (M5, parity gate); the overlay is deleted in the same commit that re-pins.(rsr-profile …)clause (string shape; keys:version :spec :declares-against :role :capabilities :preset :add :remove :policy :omitted-rationale, anything else fails closed, exactly one clause). The legacy file is read only when the deed has none..adocor.mdsatisfies, never both required. Default.adoc;(bot :name rhodibot (config :community-doc-format md))makes.mdthe named form. A bad value is an error, not a silent default. Bothcanonandcheck.wellknown-served, "Deployed site serves its.well-known/over HTTPS") is a deployment question the canon reader had turned into a file check. It is listed inoverlay::MANUAL_ONLY(recorded inpin.toml) and reported as a visiblemanualfinding, "manual check: not verified by rhodibot", with the path it is about (.well-known/). It is not graded, never trips--fail-on, and is never silently dropped. It is an explicit list, not a "detect = manual" rule, because 1.2.4 (.tool-versions) isdetect = manualyet a real file question.scripts/fleet-report-back.sh: the rhodibot translation maps severitymanualtoinfo.check(rsr.rs), fleet rule ids/suggestions and the webhook checklist drop.a2ml/.scmand require<repo>_chora.deed..claude/CLAUDE.mdinvariant 1 rewritten: repo facts are deeds, no.a2mlshould exist.Field shapes were agreed with A7 (the
:community-doc-formatrow goes into their mapping spec §6). 3.1.1 (descriptiles/) is kept: it is the home of the descriptile deeds.Proof: the real
rsr-template-repomain (563397a), via the fleet report-back driverBefore = gitbot-fleet main 14892a5; after = this branch (re-verified at 0e83966; d17c995 adds only the path on manual findings, to clear the driver's "finding with no path" warning).
.a2mldemands (0-AI-MANIFEST, 7 descriptiles, rsr-profile)CONTRIBUTING.md/CODE_OF_CONDUCT.md.tool-versions.well-known/(9.1.4).github/GOVERNANCE.adoc(2.1.8)www/.well-known/security.txt(2.2.1)Why 9.1.4 and 2.1.8 now appear
The template declares
docs-siteandgovernance-tierin the(rsr-profile …)clause of its chora deed. The old code only looked forrsr-profile.a2ml, so it never saw them and every gated criterion was silentlyna. Reading the profile brings those two criteria into scope. 9.1.4 is shown as manual info (above); 2.1.8 expectsdocs/GOVERNANCE.adocand the template keeps it under.github/, info only.Tests
cargo testinbots/rhodibotpasses (168 + 11 + 58 + 1);cargo fmt --checkandcargo clippy --all-targetsclean. Tests cover both document forms, both-present, the override, a bad override value, no A2ML in any expectation, deed-clause and descriptile satisfaction, the legacy fallback, a verbatim copy of standards' own(rsr-profile …)clause, manual findings (visible, pathed, not graded), and the pin/overlay lockstep.Follow-ups
rsr-profile.abnf: A7 will add a clause module for(rsr-profile …)(after neurosym), so one grammar owns its keys. The key set insrc/deed.rsis interim and should later come from, or be tested against, that module.src/canon/overlay.rsand the[overlay]table in the same commit that re-pins.Not done here
FLEET_REFin rsr-template-repofork-governance.ymlafter this merges (pins belong to Tsunami-breaker).🤖 Generated with Claude Code
https://claude.ai/code/session_0136P7ZmM7wgb8pXfyFEuRHQ
Summary by CodeRabbit
New Features
Documentation