Skip to content

Fix agy 1.3.3 prompt argument order - #661

Merged
JacobStephens2 merged 15 commits into
mainfrom
issue-658
Oct 12, 2026
Merged

JacobStephens2 merged 15 commits into
mainfrom
issue-658

Conversation

@JacobStephens2

@JacobStephens2 JacobStephens2 commented Oct 12, 2026 •

Copy link
Copy Markdown
Owner

Summary

-agy -p --dangerously-skip-permissions ... [--conversation <id>] <prompt>
+agy -p <skill-loaded prompt> --dangerously-skip-permissions ... [--conversation <id>]

New sessions and Resumes pass the Session prompt as -p’s value. Release summaries still use -p= with stream-json stdin. Fake agy now parses and validates the same prompt position.

Closes #658

Evidence

  • Before: cargo test --bin thirdshift harness::agy::tests::a_resume_reuses_every_flag_and_environment_and_the_conversation_id -- --exact failed on the old argument order. The new-session Command regression failed with -p took "--dangerously-skip-permissions" as its prompt.
    After: the Resume test and all nine cargo test --test agy_sessions tests pass, including the skill-loaded prompt order.

  • Before: moving the Session prompt broke cargo test --test release every_harness_receives_the_complete_release_input_above_the_argument_limit -- --exact for agy with Argument list too long.
    After: that check and every_other_harness_summarizes_without_its_session_permission_bypass pass after updating the summary's prompt removal.

  • Review-report fixture: cargo test --bin thirdshift session::execution_tests::keeps_review_reports_beside_each_harness_session_log -- --exact failed for agy with the old final-argument assumption and passes after reading agy’s -p value.

  • Live: agy 1.3.3 with gemini-3.8-flash-low, through thirdshift's production Adapter, process owner and Interpretation, returned AGY_NEW_658 and AGY_RESUMED_658, both on conversation ecf33305-7cd8-4050-b8a5-ca7a4d6c3a89. Explicit rerun passed: cargo test --bin thirdshift harness::agy::tests::real_agy_receives_new_and_resumed_prompts -- --exact --ignored --nocapture.

  • Final checks: cargo fmt --check, cargo clippy --all-targets -- -D warnings, and the full cargo test suite passed before integrating the remote branch. After integration, Clippy, both invocation/report unit checks, all nine agy Command tests, and all 52 release tests passed.

Test seams

  • Harness Adapter::session invocation for new sessions and Resumes, including Factory skill loading; live agy uses the same seam.
  • Command sessions through the compiled thirdshift CLI and fake agy.
  • Release-summary input through the compiled thirdshift CLI.
  • Sessions review-report preservation through fake Harness CLIs.

Review

Standards: one low-severity Duplicated Code finding addressed by reusing the parsed fake prompt. Spec: no findings. Both axes read every changed file; no changed files were left unread.

Unaddressed findings

Standards

None from their commits since 8fa1ee01eb4320d5afeeca82c94589da716a59f2 (final reviewed head 7af3d15a32465a731dd457d6342ec6cd69f8dc36): no findings were skipped. All four changed files were read, including site/index.html added by the concurrent merge of main; no changed files were left unread.

For their commits since 08529e767aff51ed0640607ffb7cb70e4bfa2675 (examined head 7af3d15a32465a731dd457d6342ec6cd69f8dc36), the pinned diff is empty and both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight; no reviewer was spawned, no findings were skipped, and no changed files were left unread.

For their commits since 7af3d15a32465a731dd457d6342ec6cd69f8dc36 (examined head a21bdca43112c741ba02570897718c526d42d6ff), git diff --exit-code 7af3d15a32465a731dd457d6342ec6cd69f8dc36...HEAD passed with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-hqeAyN/.

For their commits since a21bdca43112c741ba02570897718c526d42d6ff (examined head 9b98129e066a91a95c40e6e0dfbeab5b92bdc3ee), git diff --exit-code a21bdca43112c741ba02570897718c526d42d6ff...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-Frqi2w/.

For their commits since 9b98129e066a91a95c40e6e0dfbeab5b92bdc3ee (examined head 7ef69203f64dabecdd63ae4b413d3e6bb09c6093), git diff --exit-code 9b98129e066a91a95c40e6e0dfbeab5b92bdc3ee...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-sBwQTO/.

For their commits since 7ef69203f64dabecdd63ae4b413d3e6bb09c6093 (examined head 56e01f0ce593a108bbc7705b0618ea6d56b64d34), git diff --exit-code 7ef69203f64dabecdd63ae4b413d3e6bb09c6093...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-NJTV9t/.

For their commits since 56e01f0ce593a108bbc7705b0618ea6d56b64d34 (examined head 03795871ea900a9e5e2017093c6e5242e8822e38), git diff --exit-code 56e01f0ce593a108bbc7705b0618ea6d56b64d34...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-ODrvow/.

Spec

None from their commits since 8fa1ee01eb4320d5afeeca82c94589da716a59f2 (final reviewed head 7af3d15a32465a731dd457d6342ec6cd69f8dc36): no findings were skipped. All four changed files were read, including site/index.html added by the concurrent merge of main; no changed files were left unread.

For their commits since 08529e767aff51ed0640607ffb7cb70e4bfa2675 (examined head 7af3d15a32465a731dd457d6342ec6cd69f8dc36), the pinned diff is empty and both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight; no reviewer was spawned, no findings were skipped, and no changed files were left unread.

For their commits since 7af3d15a32465a731dd457d6342ec6cd69f8dc36 (examined head a21bdca43112c741ba02570897718c526d42d6ff), git diff --exit-code 7af3d15a32465a731dd457d6342ec6cd69f8dc36...HEAD passed with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-hqeAyN/.

For their commits since a21bdca43112c741ba02570897718c526d42d6ff (examined head 9b98129e066a91a95c40e6e0dfbeab5b92bdc3ee), git diff --exit-code a21bdca43112c741ba02570897718c526d42d6ff...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-Frqi2w/.

For their commits since 9b98129e066a91a95c40e6e0dfbeab5b92bdc3ee (examined head 7ef69203f64dabecdd63ae4b413d3e6bb09c6093), git diff --exit-code 9b98129e066a91a95c40e6e0dfbeab5b92bdc3ee...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-sBwQTO/.

For their commits since 7ef69203f64dabecdd63ae4b413d3e6bb09c6093 (examined head 56e01f0ce593a108bbc7705b0618ea6d56b64d34), git diff --exit-code 7ef69203f64dabecdd63ae4b413d3e6bb09c6093...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-NJTV9t/.

For their commits since 56e01f0ce593a108bbc7705b0618ea6d56b64d34 (examined head 03795871ea900a9e5e2017093c6e5242e8822e38), git diff --exit-code 56e01f0ce593a108bbc7705b0618ea6d56b64d34...HEAD exited 0 with no output and the changed-file list is empty; both trees are d8a070e82a1d100ff1f35845711f8c4c74f382f2. The review skill stopped at its empty-diff preflight. No independent reviewer was spawned, no findings were produced or skipped, and no changed files were left unread. Both axis preflight records are saved in .thirdshift-review-ODrvow/.

Merge Danger

Door: two-way

Revert restores the previous argument construction.

Blast Radius: agy

Affects agy Session prompt transport and derived release-summary arguments.

Built with codex · gpt-6.1-sol · high

CI notes

Review ebcdf14 from fixed point 8fa1ee0. Independent Standards and Spec reviews found no findings and read all three changed files. Saved the reviewer reports and updated PR #661 Unaddressed findings without changing other sections.

Formatting, Clippy, agy session, release and report checks passed. Broad suite and remaining binaries passed except the deployment verifier exposed an unrelated host systemd unit warning; its exact isolated rerun passed with SYSTEMD_UNIT_PATH pointing to repository and vendor units. Installed agy is 1.2.10, so the paid 1.3.3 live check was not repeated. No code changes were required.
The requested three-dot diff from 08529e7 to 7af3d15 is empty; both revisions have tree d8a070e. Per the review skill preflight, no independent reviewers were spawned and there were no findings to reproduce or address. Saved both axis records to .thirdshift-review-ehAyFN and appended the empty-diff evidence only to PR #661 Unaddressed findings. No changed files were left unread. Reports remain in the repository-excluded local review directory.
The requested three-dot diff from 7af3d15 to a21bdca is empty. Both revisions have tree d8a070e. The thirdshift-code-review skill stopped at its empty-diff preflight, so no independent reviewers were spawned and no behavioral findings required reproduction. Saved both axis preflight records to .thirdshift-review-hqeAyN/standards.md and spec.md, with final reviewer files-read lists and no unread changed files. These local report files are excluded by repository configuration. Appended the precise empty-diff evidence under both axes in PR #661 Unaddressed findings while preserving the rest of the body.
The requested comparison a21bdca...9b98129 is empty; both trees are d8a070e. The review skill stopped at its empty-diff preflight, so no independent reviewers were spawned and no behavioral findings required reproduction. Saved both axis preflight reports with final files-read lists and no unread changed files in .thirdshift-review-Frqi2w/. Report files remain local under the repository's existing exclusion. Appended evidence under both axes in PR #661 Unaddressed findings and verified that all existing body text was preserved.
The requested comparison 9b98129...7ef6920 is empty; both trees are d8a070e. The review skill stopped at its empty-diff preflight, so no independent reviewers were spawned and no behavioral findings required reproduction. Saved both axis preflight reports with final files-read lists and no unread changed files in .thirdshift-review-sBwQTO/. Report files remain local under the repository existing exclusion. Appended evidence under both axes in PR #661 Unaddressed findings and verified that all existing body text was preserved.
The requested comparison 7ef6920...56e01f0 is empty; both trees are d8a070e. The review skill stopped at its empty-diff preflight, so no independent reviewers were spawned and no behavioral findings required reproduction. Saved both axis preflight reports with final files-read lists and no unread changed files in .thirdshift-review-NJTV9t/. Report files remain local under the existing repository exclusion. Appended evidence under both axes in PR #661 Unaddressed findings and verified that all existing body text was preserved.
@JacobStephens2
JacobStephens2 merged commit cfda566 into main Oct 12, 2026
17 checks passed
@JacobStephens2
JacobStephens2 deleted the issue-658 branch October 12, 2026 04:05
@JacobStephens2
JacobStephens2 restored the issue-658 branch October 12, 2026 04:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

agy sessions fail on agy 1.3.3: pass the prompt as -p's value

1 participant