Merged
Conversation
Captures the v2 design for ce:ideate that expands the skill beyond repo-grounded ideation to also support elsewhere mode (greenfield product, business model, design/UX/naming, personal decisions). Key v2 additions documented: - Two-decision mode classifier (repo vs elsewhere; software vs not) with active confirmation on ambiguity - New web-researcher agent (Sonnet, phased search budget) reused across modes, with sidecar cache for session-scoped reuse - Two new always-on frames: cross-domain analogy, constraint-flipping - Mode-determined persistence defaults (Proof for elsewhere, docs/ideation/ for repo); terminal-first opt-in save - Orchestrator-side single best-effort retry around proof skill, with explicit fallback menu on persistent failure - Two auto-compact checkpoints (after Phase 2 raw candidates, before Phase 4 survivors) to bound mid-session loss
Adds a research agent that returns structured external grounding (prior art, adjacent solutions, market signals, cross-domain analogies) for use by ideation and planning skills.
Upgrades ce:ideate to classify topic mode (repo-grounded vs elsewhere; software vs non-software) and route to the appropriate grounding and facilitation path. Introduces six mode-symmetric generation frames, an adversarial critique with survivor rubric, a session-scoped web-research cache, and a terminal-first review loop so refinement happens in conversation before any persistence.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a5c63738f6
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".
- Narrow elsewhere-non-software routing so Phase 1 learnings + web-research still run - Use fully-qualified compound-engineering:research:web-researcher agent identifier - Preserve per-run scratch directory so V15 cache reuse works across run-ids
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 22b48a9617
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".
- Add caller-aware Proof return rule so §6.1 Brainstorm continues into ce:brainstorm after a successful Proof save instead of bouncing to the Phase 6 menu - Carry Phase 6.5 Proof Failure Ladder forward into elsewhere-non-software runs; wire universal-ideation.md Save-and-end to the fallback so local-save recovery stays reachable
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 78b980474f
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".
- Rewrite §5.2 caller-aware return rule to branch on §6.1 (continue into brainstorm), §6.3 (exit cleanly), and default (menu) - Update §6.3 to end cleanly after successful Proof or file save instead of looping back to the Phase 6 menu
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3c9f59dd4f
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".
- universal-ideation.md: add Grounding freshness rule to re-dispatch Phase 1 agents when intake materially refines topic (mirrors SKILL.md Phase 0.4) - universal-ideation.md: expand Brainstorm option to persist ideation per Phase 6.1 contract (save + Explored marker + caller-aware return + Proof failure ladder) before handing off to ce:brainstorm Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… Proof - Reorder wrap-up options: Refine (1) → Open and iterate in Proof (2) → Brainstorm (3) → Save and end (4). Refine is cheapest/reversible; Save and end is terminal. - Restore Open and iterate in Proof as a menu option. This invokes §5.2 HITL review (save to Proof + iterate via comments + sync reviewed edits back to docs/ideation/ in repo mode). The v2 redesign had folded this into Save and end; it was a useful standalone path that users relied on. - Drop End in conversation only to stay within the 4-option AskUserQuestion cap. Refine covers the walk-away path in practice; the free-text escape hatch remains. - Renumber subsections §6.1-§6.4 to match the new menu order; §6.5 Proof Failure Ladder unchanged. Update §5.2 caller-aware return rules accordingly (§6.2 returns to menu; §6.3 continues to brainstorm handoff; §6.4 exits cleanly). - Mirror the reorder and § references in references/universal-ideation.md for the elsewhere-non-software path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 736b3af8ee
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".
…t menu improvisation
Agents were glossing over the "read references/post-ideation-workflow.md" line and improvising the Phase 6 menu from memory, producing options ("Something else", "Refine the list") that aren't in the spec. Same risk on the elsewhere-non-software routing to universal-ideation.md.
Both load instructions now explicitly state:
- the load is non-optional
- what lives in the reference file that does not live in SKILL.md (specific menu options, frames, rubric)
- the failure mode (improvising from memory degrades subsequent steps)
- defense against speed-motivated skipping ("'quickly' means fewer Phase 2 sub-agents, not skipping references")
No behavior change when the reference is loaded; this change targets the skip-the-load failure mode specifically.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…n non-software mode, de-jargon user-facing copy
Three linked fixes for observed misbehavior when ideating about a landing page from a different system:
1. Decision 2 (software vs non-software) now classifies by subject, not by where the ideas will land. Topics about a software artifact (page, app, feature, flow, product) are elsewhere-software even when the ideas are copy, UX, CRO, pricing, or visual-design work. Non-software is reserved for topics with no software surface at all (naming, narrative, personal, non-digital business). Added disambiguating examples.
2. Learnings-researcher is now skipped by default in elsewhere-non-software. The CWD repo's docs/solutions/ almost always contains engineering patterns that do not transfer to naming/narrative/personal/non-digital-business topics; running it wastes tokens and risks polluting generation. Updated Phase 1 dispatch, Cost Transparency examples (now 8 agents in non-software default), and the universal-ideation.md header.
3. User-facing mode statement no longer leaks internal taxonomy labels. The template now uses plain language ("Treating this as a topic in this codebase", "...as a product/software topic outside this repo", "...as a [naming|narrative|business|personal] topic") with plain-language correction hints ("actually this is outside the repo", "actually this is about a software product"), never "elsewhere-non-software" or "repo-grounded".
Together these make ideation of a sign-up page in an unrelated CWD both classify correctly (elsewhere-software, so software frames run) and narrate clearly to the user.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…nvention to default OS temp
AGENTS.md Scratch Space rule is inverted. The previous rule defaulted scratch to .context/ and treated OS temp as the exception; it caused ce-ideate to pollute whatever CWD happened to be, even when the ideation topic was unrelated to that repo (e.g., ideating about another system's landing page while CWD was image-sprout).
New rule:
- Default: OS temp — covers most scratch, including per-run throwaway (mktemp -d) AND cross-invocation reusable (stable "${TMPDIR:-/tmp}/compound-engineering/<skill>/<run-id>/" prefix). Regardless of whether a repo is present or whether other skills may read the files.
- Exception: .context/ — only when the artifact is genuinely bound to the CWD repo AND meets one of: user-curated (TODO database, optimization log); repo+branch-inseparable (branch-specific resume state); or path-is-core-UX (surfaced back to user as part of skill output).
- "Shared between skills" alone is not sufficient justification — stable OS-temp prefixes handle coordination equally well.
Migrations applied:
- ce-ideate: checkpoints A/B and V15 web-research cache now live in "${TMPDIR:-/tmp}/compound-engineering/ce-ideate/<run-id>/". Added explicit Phase 1 pre-resolution step so the agent captures the absolute path once via bash and uses it for all Write/Read calls (unresolved shell strings would break with non-shell tools).
- ce-work-beta codex-delegation: per-run throwaway, switched to mktemp -d. Dropped explicit rm -rf cleanup since OS handles it; leaving files preserves artifacts for debugging.
- ce-plan deepen (artifact-backed mode): per-run throwaway, switched to mktemp -d. Sub-agents receive the absolute path explicitly. Dropped explicit cleanup.
Kept on .context/: ce-optimize (per-spec persistent state across sessions; user-curated), ce-review (path is surfaced as core UX), todo-create/triage/resolve todos/ database (user-manipulated per-repo TODOs).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- ce-ideate SKILL.md: make Phase 2 dispatch count mode-conditional (4 sub-agents in issue-tracker mode, 6 otherwise) to resolve the contradiction between the Phase 2 intro and the issue-tracker override at line 254 - ce-ideate SKILL.md: fix Phase 0.5 Cost Transparency arithmetic — issue-tracker mode trades 6 ideation for 4 AND adds 1 issue-intelligence agent, so the net is -1 (baseline 8), not +1 - ce-ideate SKILL.md: add explicit "Repo mode, issue-tracker intent" example in the cost-transparency list Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3cdef56be1
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".
- ce-work-beta codex-delegation: replace $SCRATCH_DIR shell-variable references in Step A codex-exec, Step B poll, and cleanup blocks with the <scratch-dir> placeholder convention used elsewhere in the workflow. Separate Bash tool calls start fresh shells, so a shell-local var from setup would expand empty and break result detection. Added explicit substitution notes above the affected blocks. - ce-ideate Phase 0.2 Routing Rule: align the elsewhere-non-software grounding narration with the Phase 1 default (user-context synthesis + web-research; learnings skipped for non-software). Previous wording still listed learnings as part of the default dispatch, contradicting the skip rule added in an earlier round. - universal-ideation.md Grounding freshness: fix the stale "user-context synthesis + learnings + web-research" listing to match the new non-software default. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 87988fa3b9
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".
…here mode can end naturally The default flow was biased toward "ideate -> brainstorm -> plan -> implement," which only fits repo mode. In elsewhere modes the user is typically done after ideation (possibly + Proof iteration); there is no ce:plan -> ce:work chain downstream, and forcing continuation produces decision fatigue after the user already got what they came for. Three linked changes: 1. Scope §6.3's "don't skip brainstorming and go straight to planning" guardrail to repo mode. In elsewhere modes, ideation (or ideation + Proof iteration) is a legitimate terminal state; brainstorming is optional deeper development of one idea. 2. Make §5.2 + §6.2 post-Proof behavior mode-aware. Repo mode returns to the Phase 6 menu on every status as before (user may still want to brainstorm toward a plan or keep refining in the repo). Elsewhere mode exits cleanly on successful Proof return — the artifact lives at docUrl, and forcing another menu choice is friction. Only aborted returns to the menu in both modes. 3. Rewrite the Brainstorm option description in universal-ideation.md to set correct expectations: "go deeper on one idea through dialogue," with an explicit note that this is not the first step of an implementation chain and ce:brainstorm in universal mode develops the idea into a richer artifact (brand brief, plot outline, decision framework) and ends there. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2e51dccb51
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".
- web-research-cache.md: replace ls glob with find to avoid non-zero exit on first-run (no cache files). find exits 0 with empty output on no match, so the reuse-check step no longer risks aborting Phase 1. - web-research-cache.md: differentiate cache keys by full mode variant (repo | elsewhere-software | elsewhere-non-software). Cross-sub-mode reuse was possible because both elsewhere variants collapsed to a single "elsewhere" key; reclassification between them on the same topic hash would hit stale cross-domain cache. - post-ideation-workflow.md + universal-ideation.md: make the no-save exit path legible from the menu itself without adding a 5th option. Refine's label now explicitly says "(or stop here - no save)", a note below each 4-option list points out the two no-save paths (pick Refine and stop, or use the question tool's free-text escape), and §6.1 Refine notes that ending after (or without) refinement is a valid no-save exit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cba78af0b1
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".
- web-research-cache.md: repo-mode cache key discriminator is now a hashed, stable identifier — `git remote get-url origin` → `git rev-parse --show-toplevel` → cwd, hashed to a short sha256 prefix. Basename-only identifiers collided across unrelated repos sharing names (app, frontend) under the shared OS-temp scratch root. Clarified that normalization applies to focus hint and topic text, not the raw discriminator input. - ce-ideate SKILL.md: Phase 2 dispatch now gates the 4-agent issue-tracker mode on both issue-tracker intent AND usable themes returned by issue-intelligence. When intent triggered but signal was insufficient (Phase 1 insufficient-issue-signal fallback), dispatch the 6-agent default. Updated the Phase 0.5 cost-transparency example to note the insufficient-signal fallback case. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Merged
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: da89df04cf
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".
|
|
||
| If the user confirms or selects "elsewhere," still run Decision 2 to choose between elsewhere-software and elsewhere-non-software. | ||
|
|
||
| **Routing rule.** When Decision 2 = non-software, still run Phase 1 Elsewhere-mode grounding (user-context synthesis + web-research by default; skip phrases honored). Learnings-researcher is skipped by default in this mode — the CWD's `docs/solutions/` rarely transfers to naming, narrative, personal, or non-digital business topics; see Phase 1 for the full rationale. Then load `references/universal-ideation.md` and follow it in place of Phase 2's software frame dispatch and the Phase 6 menu narrative. This load is non-optional — the file contains the domain-agnostic generation frames, critique rubric, and wrap-up menu that replace Phase 2 and the post-ideation menu for this mode, and none of those details live in this main body. Improvising from memory produces the wrong facilitation for non-software topics. Do not run the repo-specific codebase scan at any point. The §6.5 Proof Failure Ladder in `references/post-ideation-workflow.md` still applies — load and follow it whenever a Proof save (the elsewhere-mode default for Save and end) fails, so the local-save fallback path stays reachable in non-software elsewhere runs. |
There was a problem hiding this comment.
This new non-software rule forbids running the repo scan, but the skill still has a global core principle that says to scan the codebase first for ideation runs; the two directives now conflict and can cause mode-misrouted behavior (either unnecessary repo scanning for elsewhere topics or skipping intended non-software flow). Please make the scan requirement explicitly mode-conditional so elsewhere runs have a single unambiguous instruction set.
Useful? React with 👍 / 👎.
Comment on lines
+28
to
+29
| SCRATCH_ROOT="${TMPDIR:-/tmp}/compound-engineering/ce-ideate" | ||
| find "$SCRATCH_ROOT" -maxdepth 2 -name 'web-research-cache.json' -type f 2>/dev/null |
There was a problem hiding this comment.
The reuse lookup scans every run directory under a shared OS-temp root without any session discriminator, but the workflow describes this as session-scoped reuse; this means a later invocation can silently reuse stale research from an older session when keys match, skipping fresh web grounding. Add a session partition (path/key) or age gate before accepting a cached match as reusable.
Useful? React with 👍 / 👎.
tmchow
added a commit
that referenced
this pull request
Apr 17, 2026
Upstream PR #588 added mode-aware v2 ideation with a new web-researcher agent and restructured post-ideation-workflow. Files included several stale refs from the pre-rename era: - agents/research/web-researcher.md was introduced without ce- prefix; renamed file to ce-web-researcher.md, updated frontmatter name, and fixed its in-prose skill refs (ce:brainstorm, ce:plan, and the stray compound-engineering:ce-ideate) to match our naming convention - ce-ideate/SKILL.md: /ce:ideate -> /ce-ideate, and all compound- engineering:research:<agent> refs simplified to <category>:ce-<agent> - ce-ideate/references/universal-ideation.md: ce:ideate, ce:brainstorm, ce:plan, ce:work refs updated to hyphenated form - README.md: ce-web-researcher added to Research agent table alongside our existing ce- prefixed agents Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
tmchow
added a commit
that referenced
this pull request
Apr 18, 2026
Upstream PR #588 added mode-aware v2 ideation with a new web-researcher agent and restructured post-ideation-workflow. Files included several stale refs from the pre-rename era: - agents/research/web-researcher.md was introduced without ce- prefix; renamed file to ce-web-researcher.md, updated frontmatter name, and fixed its in-prose skill refs (ce:brainstorm, ce:plan, and the stray compound-engineering:ce-ideate) to match our naming convention - ce-ideate/SKILL.md: /ce:ideate -> /ce-ideate, and all compound- engineering:research:<agent> refs simplified to <category>:ce-<agent> - ce-ideate/references/universal-ideation.md: ce:ideate, ce:brainstorm, ce:plan, ce:work refs updated to hyphenated form - README.md: ce-web-researcher added to Research agent table alongside our existing ce- prefixed agents Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
tmchow
added a commit
that referenced
this pull request
Apr 18, 2026
Upstream PR #588 added mode-aware v2 ideation with a new web-researcher agent and restructured post-ideation-workflow. Files included several stale refs from the pre-rename era: - agents/research/web-researcher.md was introduced without ce- prefix; renamed file to ce-web-researcher.md, updated frontmatter name, and fixed its in-prose skill refs (ce:brainstorm, ce:plan, and the stray compound-engineering:ce-ideate) to match our naming convention - ce-ideate/SKILL.md: /ce:ideate -> /ce-ideate, and all compound- engineering:research:<agent> refs simplified to <category>:ce-<agent> - ce-ideate/references/universal-ideation.md: ce:ideate, ce:brainstorm, ce:plan, ce:work refs updated to hyphenated form - README.md: ce-web-researcher added to Research agent table alongside our existing ce- prefixed agents Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
tmchow
added a commit
that referenced
this pull request
Apr 18, 2026
Upstream PR #588 added mode-aware v2 ideation with a new web-researcher agent and restructured post-ideation-workflow. Files included several stale refs from the pre-rename era: - agents/research/web-researcher.md was introduced without ce- prefix; renamed file to ce-web-researcher.md, updated frontmatter name, and fixed its in-prose skill refs (ce:brainstorm, ce:plan, and the stray compound-engineering:ce-ideate) to match our naming convention - ce-ideate/SKILL.md: /ce:ideate -> /ce-ideate, and all compound- engineering:research:<agent> refs simplified to <category>:ce-<agent> - ce-ideate/references/universal-ideation.md: ce:ideate, ce:brainstorm, ce:plan, ce:work refs updated to hyphenated form - README.md: ce-web-researcher added to Research agent table alongside our existing ce- prefixed agents Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
tmchow
added a commit
that referenced
this pull request
Apr 18, 2026
Upstream PR #588 added mode-aware v2 ideation with a new web-researcher agent and restructured post-ideation-workflow. Files included several stale refs from the pre-rename era: - agents/research/web-researcher.md was introduced without ce- prefix; renamed file to ce-web-researcher.md, updated frontmatter name, and fixed its in-prose skill refs (ce:brainstorm, ce:plan, and the stray compound-engineering:ce-ideate) to match our naming convention - ce-ideate/SKILL.md: /ce:ideate -> /ce-ideate, and all compound- engineering:research:<agent> refs simplified to <category>:ce-<agent> - ce-ideate/references/universal-ideation.md: ce:ideate, ce:brainstorm, ce:plan, ce:work refs updated to hyphenated form - README.md: ce-web-researcher added to Research agent table alongside our existing ce- prefixed agents Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Before:
ce:ideateonly worked well when the topic was the current codebase. Non-code prompts (naming, pricing, narrative, personal decisions) got forced through software-flavored frames and produced generic output. Scratch state polluted whatever repo happened to be in CWD, and the post-ideation menu assumed every run would continue into a software-implementation chain.After:
ce:ideateclassifies the subject of ideation in two steps (repo-grounded vs elsewhere, then software vs non-software) and routes each mode to the right facilitation path. Six mode-symmetric frames replace the software-only set. Persistence, brainstorm handoff, and post-Proof behavior are all mode-aware — elsewhere ideation can terminate cleanly without forcing users past their natural stopping point. A new plugin-wide scratch-space rule defaults to OS temp so ideation scratch no longer lands inside an unrelated CWD repo.The new
web-researcheragent and the scratch-space convention inversion ship in the same PR because mode-aware elsewhere ideation depends on both: without external grounding there is nothing to replace the codebase scan, and without the scratch migration, writing.context/into an unrelated repo during elsewhere-mode ideation is the canonical bad UX.Mode classification
Classification runs on the subject of ideation, not the environment. A user in
/tmpcan ideate about code in their head; a user in a repo can ideate about an unrelated product's landing page.Phase 6 handoff menu
The post-ideation wrap-up offers 4 options: Refine, Open and iterate in Proof, Brainstorm, Save and end. Behavior is mode-aware:
docs/ideation/, return to menu for follow-up work.ce:brainstorm→ce:plan→ce:worktoward implementation.ce:brainstormin universal mode develops the idea into a richer artifact (brand brief, plot outline, decision framework) and ends there.docs/ideation/, exit.The §6.3 guardrail "don't skip brainstorming and go straight to planning" is scoped to repo mode. Elsewhere mode has no
ce:plan→ce:workchain downstream, so ideation (or ideation + Proof iteration) is a legitimate terminal state.External grounding
web-researcheragent (new) performs iterative web research and returns structured grounding: prior art, adjacent solutions, market signals, cross-domain analogies. Available to ideation and future skills.re-research. Cache keys differentiate by full mode variant (repo/elsewhere-software/elsewhere-non-software) so software and non-software research cannot cross-reuse on the same topic hash.docs/solutions/rarely transfers to naming, narrative, personal, or non-digital business topics.Scratch-space convention
AGENTS.mdinverts the plugin-wide Scratch Space rule: default to OS temp; use.context/only when justified. The previous rule defaulted to.context/, which caused ideation scratch to land inside whatever CWD repo the user happened to be in, even when the topic was unrelated to that repo.The new rule allows
.context/only when the artifact is genuinely bound to the CWD repo AND is user-curated, repo+branch-inseparable, or has a path that is core UX.Migrations applied:
"${TMPDIR:-/tmp}/compound-engineering/ce-ideate/<run-id>/". Phase 1 pre-resolves the absolute path once via bash so Write/Read tools always receive resolved paths.mktemp -d. Explicitrm -rfdropped; OS handles cleanup, and orphaned dirs preserve artifacts for debugging.mktemp -d. Sub-agents receive the absolute path as an argument.Kept on
.context/:ce-optimize(per-spec persistent state across sessions),ce-review(artifact path is surfaced as core UX), todo-create/triage/resolve (per-repo TODO database that users inspect directly).Design notes
references/post-ideation-workflow.md(the Phase 6 menu) andreferences/universal-ideation.md(non-software facilitation) carry load instructions that explicitly state the load is non-optional, name what lives in the reference that does not live inSKILL.md, and defend against speed-motivated skipping. Without these, agents were improvising the menu from memory.issue-intelligence-analyst, for a net of 8 agents. The Phase 0.5 cost line reflects this explicitly.Testing
Tested end-to-end by running v1 (pre-branch) and v2 (this branch) against a real repo (
~/Code/image-sprout, a Svelte + TypeScript image-generation tool) with identical prompts, plus v2 on non-software prompts and a landing-page conversion prompt:"improve this project"returned 6 survivors, 5 of which were mechanical refactor recommendations (split CLI monolith, Zod schemas, smoke tests, etc.)."vision-scale product expansion ideas"for the same repo produced genuine product reframings (Brand-Kit Mode, Evaluator-Loop Mode, MCP-server mode, Guide Version Control, pluggable output destinations).No test suite changes (skill content is markdown).
bun run release:validatepasses.