feat(aidd-context): add Kilo-native generation targets by waewoo · Pull Request #992 · ai-driven-dev/framework

waewoo · 2026-10-10T17:52:14Z

🎯 What & why

Adds native Kilo support to "aidd-context", enabling Kilo-only projects to onboard AIDD and generate discoverable project memory, rules, skills, agents, and workflows while preserving compatibility with other supported tools.

🛠️ How it works

Detection and memory: Recognizes canonical and legacy Kilo project signals without confusing them with OpenCode. Uses shared "AGENTS.md" project memory while preserving user-owned content.
Rules: Publishes ".kilo/rules/" artifacts and registers their exact paths in "kilo.json[c]" "instructions". The shared writer preserves existing JSONC content, handles ambiguous configurations explicitly, and uses preflight checks, staging, and best-effort rollback.
Skills: Supports native ".kilo/skills/" placement and explicit portable ".agents/skills/" placement.
Agents and workflows: Generates native Markdown artifacts under ".kilo/agents/" and ".kilo/commands/", using documented fields.
Hooks: Provides Kilo JavaScript/TypeScript plugin guidance without generating unsupported declarative hooks. Existing host contracts remain supported.
Cross-tool improvements: Hardens shared memory filesystem handling and rule publishing, strengthens hook compatibility checks, and extends "skill-eval" with Claude/Codex evaluation support, JSON file checks, project-tree snapshots, and explicit "PASS"/"FAIL"/"INCOMPLETE" reporting.

Detailed host contracts are documented in "aidd-context" (https://github.com/ai-driven-dev/framework/blob/ccd32e3bd9075cce43be6bea4a7bb1c60a14e6dd/plugins/aidd-context/README.md) and its generator references.

🧪 How to verify

CI passes on the current PR HEAD, including Windows CLI checks, mutation tests, CodeQL, and Kilo runtime E2E tests.

For focused Kilo validation:

cd cli
pnpm test:e2e:kilo
pnpm test:e2e:kilo:native

The CI runtime tests use Kilo CLI 7.7.5 with a deterministic local provider on Linux.

Additionally, phase 6 records a successful authenticated Codex CLI 0.162.1 execution using AIDD skills to generate a native Kilo skill, its action, and a synthetic payload. The generated artifacts passed the scoped case assertions.

A subsequent Kilo 7.8.8 Linux runtime test verified discovery, skill loading, action reading, and payload consumption using a deterministic local provider.

These are scoped behavioral and runtime validations, not evidence that the entire "skill-eval --host=codex" suite was executed successfully.

⚠️ Heads-up

Claude validation: An authenticated Claude Code session has not been performed. "CONTRIBUTING.md" requests testing with Claude and another tool; maintainer guidance is requested on accepting the available Codex/Kilo evidence or arranging a Claude validation.
Platform coverage: Kilo runtime behavior has been validated on Linux. Windows CI covers the general CLI, not authenticated or native Kilo runtime execution on Windows. Kilo runtime validation on Windows/macOS remains outstanding.
Filesystem safety: Shared project-memory writes and rule publication have been hardened, but they do not guarantee atomic multi-file updates or recovery from every filesystem failure.
Manual memory synchronization: The manual workflow provides fewer filesystem protections than the executable hook. Generic installed-plugin resource resolution is tracked separately in "#994" (feat(framework): resolve resources shipped with installed plugins #994).
Partial writes: "#1018" (fix(aidd-context): preserve project-memory files on write failure #1018) tracks a project-memory corruption bug reproduced on "next" when an I/O failure occurs during writing. This PR does not claim to resolve that issue.
Skill evaluation: Results marked "INCOMPLETE" do not establish a successful LLM-judged evaluation, even when the process exits with code "0".
Detection coverage (AC1–4): Explicit detection guidance is incomplete for rule and hook generators; no incorrect detection was reproduced.
Generation guarantees (AC13–14): Idempotence and safe-write behavior are covered by targeted tests, but complete guarantees for every LLM-guided generation path remain partially verified.
Official Kilo references (AC16): Official Kilo paths, formats, and sources are documented centrally in "official-sources.md" (https://github.com/ai-driven-dev/framework/blob/ccd32e3bd9075cce43be6bea4a7bb1c60a14e6dd/aidd_docs/tasks/2026_10/2026_10_10_kilo-native-generation-914/official-sources.md). Issue feat(aidd-context): add Kilo-native generation targets #914 records a verification date of September 25, 2026; the documentation was independently rechecked on October 13, 2026. Maintainer confirmation is requested on whether this centralized approach satisfies the per-claim traceability requirement.

🔗 Linked issue

Refs #914

Follow-up issues

Three issues are tracked separately:

feat(framework): resolve resources shipped with installed plugins #994: Reliable resolution of installed plugin resources. The manual memory-sync workflow does not provide the executable hook's full protections.
fix(aidd-context): preserve project-memory files on write failure #1018: Partial writes can corrupt project-memory files on I/O failure. Reproduced on next; not fully addressed by this PR's filesystem hardening.
fix(cli): isolate ambient session IDs in telemetry lifecycle tests #1050: Ambient CODEX_THREAD_ID can override the test fixture's CLAUDE_CODE_SESSION_ID, causing two telemetry lifecycle tests to fail. Tracked separately; this PR does not claim to resolve it.

None of these issues is claimed as resolved here.

✅ I certify

I DO CERTIFY I READ EACH LINE OF THE PULL REQUEST BECAUSE I AM A SOFTWARE ENGINEER, NOT A AI PUPPY.

Cover six canonical and legacy signals without false Kilo detection from shared AGENTS or OpenCode config. Keep OpenCode detection unchanged. TASK 1 of ai-driven-dev#914 phase 1. Detection contracts: 24/24 pass; removing the nested JSONC signal causes a targeted failure.

Reuse the shared AGENTS destination and preflight all selected files before Upsert and Fill. Preserve user bytes and keep automatic hooks best-effort. Reject known unsafe destinations before any selected write. TASK 2 of ai-driven-dev#914 phase 1. 40 memory tests and 24 detection tests pass. Final make check: 604 pass, 2 existing skips, 0 fail. Real Kilo reads linked memory and applies its instruction at reported cost zero.

Record TASK 3 runtime proof, review, assertions, task statuses and resumption decisions for ai-driven-dev#914 phase 1. Preserve the approved plan and its previous snapshot; later phases remain pending and require separate authorization. Native Codex generated shared memory. Real Kilo read the linked file and applied its instruction at reported cost zero. Final checks: 64 targeted pass; 604 full-script pass, 2 existing skips; review approve. Claude authenticated runtime and standalone Codex CLI limitations remain explicit.

Require an explicit portable choice and resolve shared target conflicts before publication. Preserve host field contracts and user resources with a full preflight, and record actual caller generation and Kilo runtime evidence. Scope is limited to validated phase 3. The overall plan remains in progress; Claude authentication, other Kilo platforms and subsequent phases remain open. Refs ai-driven-dev#914

Implement phase 4 of ai-driven-dev#914 with Kilo contracts, regression coverage, and Linux runtime proof.

docs: archive phase 5 validation reruns Record targeted Persona and pinned Kilo runtime evidence for ai-driven-dev#914. test(aidd-context): prove hook contract assertions Exercise seven targeted source-contract mutations for ai-driven-dev#914 and tighten the two assertions that survived initial probes. Archive final phase-5 validation and update the coverage matrix and review.

Verify project-file descriptors while preserving user content and metadata, and refuse detected substitutions and hard-linked targets. Use private test fixtures and descriptor-backed witnesses. Make line-ending, permission, mode, and parent-identity expectations explicit across platforms. Refs ai-driven-dev#914

waewoo · 2026-10-10T22:22:37Z

Quick progress update on #992:

Phases 1, 3, 4, and 5 are complete, with CI passing.
Phase 2 remains partially implemented, pending coordination with fix(aidd-context): generate rules through a standalone skill script #979 to reuse the shared rule writer.
Phase 6 is still pending. Authenticated Claude runtime validation is currently unavailable in my environment, so I’m not claiming that coverage.
The memory-sync workflow now uses a constrained manual process, without claiming the executable hook's hardlink and file-identity protections.

I've opened #994 to track the underlying plugin resource-resolution limitation separately, without expanding the scope of this PR.

I'll keep #992 in Draft until the remaining work is addressed.

Cover six canonical and legacy signals without false Kilo detection from shared AGENTS or OpenCode config. Keep OpenCode detection unchanged. TASK 1 of ai-driven-dev#914 phase 1. Detection contracts: 24/24 pass; removing the nested JSONC signal causes a targeted failure.

Reuse the shared AGENTS destination and preflight all selected files before Upsert and Fill. Preserve user bytes and keep automatic hooks best-effort. Reject known unsafe destinations before any selected write. TASK 2 of ai-driven-dev#914 phase 1. 40 memory tests and 24 detection tests pass. Final make check: 604 pass, 2 existing skips, 0 fail. Real Kilo reads linked memory and applies its instruction at reported cost zero.

Record TASK 3 runtime proof, review, assertions, task statuses and resumption decisions for ai-driven-dev#914 phase 1. Preserve the approved plan and its previous snapshot; later phases remain pending and require separate authorization. Native Codex generated shared memory. Real Kilo read the linked file and applied its instruction at reported cost zero. Final checks: 64 targeted pass; 604 full-script pass, 2 existing skips; review approve. Claude authenticated runtime and standalone Codex CLI limitations remain explicit.

Require an explicit portable choice and resolve shared target conflicts before publication. Preserve host field contracts and user resources with a full preflight, and record actual caller generation and Kilo runtime evidence. Scope is limited to validated phase 3. The overall plan remains in progress; Claude authentication, other Kilo platforms and subsequent phases remain open. Refs ai-driven-dev#914

Implement phase 4 of ai-driven-dev#914 with Kilo contracts, regression coverage, and Linux runtime proof.

docs: archive phase 5 validation reruns Record targeted Persona and pinned Kilo runtime evidence for ai-driven-dev#914. test(aidd-context): prove hook contract assertions Exercise seven targeted source-contract mutations for ai-driven-dev#914 and tighten the two assertions that survived initial probes. Archive final phase-5 validation and update the coverage matrix and review.

Verify project-file descriptors while preserving user content and metadata, and refuse detected substitutions and hard-linked targets. Use private test fixtures and descriptor-backed witnesses. Make line-ending, permission, mode, and parent-identity expectations explicit across platforms. Refs ai-driven-dev#914

Extend the canonical rule publication flow for Kilo while preserving project configuration and existing host behavior. Refs ai-driven-dev#914

…in guidance

…neration-914

…o-native-generation-914

…neration-914

Refs ai-driven-dev#914

Document merged Kilo instructions and cover coexistence in the real 7.7.5 runtime. Refs ai-driven-dev#914

Keep the plugin overview focused on shared behavior and avoid repeating the Kilo config-selection contract. Remove private runtime identifiers from retained task evidence. Refs ai-driven-dev#914

Document standalone detection and official Kilo sources. Add rerun and collision checks to support ai-driven-dev#914 acceptance review.

waewoo · 2026-10-14T05:42:27Z

The implementation is ready for maintainer review. The shared rule publisher is now integrated (#979 is no longer blocking), and CI is green, including Kilo 7.7.5 Linux E2E, Windows CLI, mutation tests, and CodeQL.

I'd appreciate your feedback on three points:

Cross-tool scope

Beyond Kilo-native generation, this PR includes:

Shared rule publishing with preflight, configuration preservation, and best-effort rollback.
Project-memory filesystem hardening and hook compatibility checks.
"skill-eval" extensions: Claude/Codex host support, JSON checks, snapshots, and "PASS"/"FAIL"/"INCOMPLETE" reporting.

The first changes support Kilo integration and existing host compatibility. The "skill-eval" extensions are less directly related.

Would you prefer to keep everything together or extract the independent changes into a follow-up PR?

AC16 — Kilo documentation provenance

AC16 requires an official source and the September 25, 2026 verification date for every Kilo path claim.

I centralized the references in "official-sources.md" (https://github.com/ai-driven-dev/framework/blob/ccd32e3bd9075cce43be6bea4a7bb1c60a14e6dd/aidd_docs/tasks/2026_10/2026_10_10_kilo-native-generation-914/official-sources.md), recording the actual October 13 re-verification rather than claiming independent evidence for September 25.

Does this centralized approach satisfy AC16's intent?

Claude validation

"CONTRIBUTING.md" requires testing with Claude and another tool. I don't have a Claude subscription or budget for paid API usage.

However, an authenticated Codex CLI 0.162.1 run successfully generated a native Kilo skill using AIDD, and a subsequent Kilo 7.8.8 Linux runtime test verified discovery, loading, and payload consumption. This is a scoped validation, not a full "skill-eval" suite pass.

Would you accept this Codex/Kilo evidence as an exception to the Claude requirement, or could a maintainer perform the Claude validation?

Follow-up issues

Three follow-up issues are tracked separately:

feat(framework): resolve resources shipped with installed plugins #994: Reliable resolution of installed plugin resources. The manual memory-sync workaround does not provide the executable hook's full protections.
fix(aidd-context): preserve project-memory files on write failure #1018: Partial writes can corrupt project-memory files on I/O failure. Reproduced on "next"; not fully addressed by this PR's filesystem hardening.
fix(cli): isolate ambient session IDs in telemetry lifecycle tests #1050: Ambient CODEX_THREAD_ID can override the CLAUDE_CODE_SESSION_ID fixture in two telemetry lifecycle tests. Reproduced on next at 940e8e36; tracked separately and not resolved by this PR.

None of these issues is claimed as resolved here.

Other validation limitations are documented in the PR description.

Thanks for your guidance before merging!

…neration-914 # Conflicts: # cli/tests/e2e/sandbox-reaches-no-tool-binary.e2e.test.ts # plugins/aidd-context/skills/05-rule-generate/actions/01-capture-rule.md # plugins/aidd-context/skills/05-rule-generate/actions/02-write-rule.md # plugins/aidd-context/skills/05-rule-generate/actions/03-validate.md # plugins/aidd-context/skills/05-rule-generate/references/tool-paths.md # plugins/aidd-context/skills/05-rule-generate/scripts/write-rule.cjs

waewoo added 7 commits October 10, 2026 11:00

feat(aidd-context): generate native Kilo agents and workflows

781125f

Implement phase 4 of ai-driven-dev#914 with Kilo contracts, regression coverage, and Linux runtime proof.

feat(aidd-context): prepare isolated Kilo config editing

43ab1cc

github-advanced-security Bot found potential problems Oct 10, 2026

View reviewed changes

Comment thread scripts/__tests__/update-memory.test.js Fixed

Comment thread scripts/__tests__/update-memory.test.js Fixed

Comment thread scripts/__tests__/update-memory.test.js Fixed

Comment thread scripts/__tests__/update-memory.test.js Fixed

waewoo force-pushed the feat/kilo-native-generation-914 branch from d5f1309 to bbd194a Compare October 10, 2026 20:47

fix(aidd-context): harden manual memory sync and Kilo guidance

0a3b728

waewoo mentioned this pull request Oct 10, 2026

feat(framework): resolve resources shipped with installed plugins #994

Open

9 tasks

waewoo added 16 commits October 12, 2026 07:00

feat(aidd-context): generate native Kilo agents and workflows

5c7cdcc

Implement phase 4 of ai-driven-dev#914 with Kilo contracts, regression coverage, and Linux runtime proof.

feat(aidd-context): prepare isolated Kilo config editing

9eb5790

fix(aidd-context): harden manual memory sync and Kilo guidance

5734c64

feat(aidd-context): publish Kilo rules through shared writer

35edbdc

Extend the canonical rule publication flow for Kilo while preserving project configuration and existing host behavior. Refs ai-driven-dev#914

test(aidd-context): validate Kilo native generation runtime

7a2d5d2

test(aidd-context): decouple generation tests from archived evidence

ecebec9

fix(skill-eval): report skipped judges as incomplete

349d260

docs(aidd-context): clarify Kilo hook generation paths

ab0dd08

fix(aidd-context): clarify multi-target generation guarantees

caadf14

fix(aidd-context): require documented hook equivalence and unify plug…

ea8eb07

…in guidance

waewoo added 3 commits October 12, 2026 16:21

docs(aidd-context): consolidate Kilo generation evidence

2ebdfa5

Merge remote-tracking branch 'upstream/next' into feat/kilo-native-ge…

118e941

…neration-914

Merge commit '0a3b7287225cf0b268afd1af022a6bf3900d4a13' into feat/kil…

0f2fac9

…o-native-generation-914

github-advanced-security Bot found potential problems Oct 12, 2026

View reviewed changes

Comment thread scripts/skill-eval.mjs Fixed

Comment thread scripts/__tests__/kilo-rule-publication.test.js Fixed

Comment thread scripts/__tests__/kilo-rule-publication.test.js Fixed

waewoo added 3 commits October 12, 2026 17:55

test(aidd-context): make Kilo rollback fault injection portable

1973fc1

fix(aidd-context): preserve exact file identities during memory sync

ea0769f

fix(aidd-context): harden skill evaluation snapshots and test preloads

6241f1f

github-advanced-security Bot found potential problems Oct 12, 2026

View reviewed changes

Comment thread scripts/skill-eval-snapshot.mjs Fixed

waewoo added 12 commits October 12, 2026 20:29

fix(aidd-context): avoid pre-open snapshot checks

c47c705

Merge remote-tracking branch 'upstream/next' into feat/kilo-native-ge…

2dccac7

…neration-914

test(ci): pin native kilo e2e to 7.7.5

acb139c

Refs ai-driven-dev#914

test(aidd-context): cover merged Kilo instructions configs

3c8d9d8

Document merged Kilo instructions and cover coexistence in the real 7.7.5 runtime. Refs ai-driven-dev#914

fix(aidd-context): close memory sync descriptors on all paths

ad03fd9

test(aidd-context): use native path for README injection

e75952b

docs(aidd-context): clarify evaluation outcomes and clean PR diff

f9721e2

docs(aidd-context): clarify Kilo path provenance

0696236

docs(aidd-context): present multi-tool capabilities

5a16b14

Keep the plugin overview focused on shared behavior and avoid repeating the Kilo config-selection contract. Remove private runtime identifiers from retained task evidence. Refs ai-driven-dev#914

test(aidd-context): remove redundant generation fixtures

11fa59a

test(kilo): streamline test comments to meet architecture budget

64a3852

docs(aidd-context): clarify multi-tool generation contracts

b9b31cd

waewoo mentioned this pull request Oct 13, 2026

fix(aidd-context): preserve project-memory files on write failure #1018

Open

1 task done

test(aidd-context): strengthen Kilo acceptance evidence

ccd32e3

Document standalone detection and official Kilo sources. Add rerun and collision checks to support ai-driven-dev#914 acceptance review.

waewoo marked this pull request as ready for review October 14, 2026 05:46

waewoo requested a review from a team as a code owner October 14, 2026 05:46

waewoo mentioned this pull request Oct 17, 2026

fix(cli): isolate ambient session IDs in telemetry lifecycle tests #1050

Open

1 of 7 tasks

Navigation Menu

Repository navigation

feat(aidd-context): add Kilo-native generation targets - #992

feat(aidd-context): add Kilo-native generation targets#992
waewoo wants to merge 45 commits into
ai-driven-dev:nextfrom
waewoo:feat/kilo-native-generation-914

waewoo commented Oct 10, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

waewoo commented Oct 10, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

waewoo commented Oct 14, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

waewoo commented Oct 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

waewoo commented Oct 10, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

waewoo commented Oct 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

waewoo commented Oct 10, 2026 •

edited

Loading

waewoo commented Oct 14, 2026 •

edited

Loading