Repository navigation
Open
Labels
agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.Fully specified and briefed; eligible for autonomous pickup from the frontier.area: securitySecurity-relevant: vulnerability, hardening, or disclosure follow-up.Security-relevant: vulnerability, hardening, or disclosure follow-up.needs-triageNot yet classified. Floor until a type and one priority tier are set.Not yet classified. Floor until a type and one priority tier are set.
Description
Parent
Refs #15703
Source: PLAN Phase 11, topic ecosystem-specifics
Depends on
- Run only stated or inferred commands unattended and report the unresolved cell #15731
- Delete toolchain's bundled ecosystem defaults #15733
Outcome
A lane's proposed ecosystem record contains only commands taken from that run's assumed: lines; any other command makes the writer exit non-zero and write nothing. A consumer can run the same check in its own CI.
Demo: feed the writer three assumed: lines and get a record with exactly those three values; add a fourth command that is not in the lines and the writer exits non-zero and writes nothing.
This item bounds what a lane may record; its review gate is the security review, run at the frontier tier.
What is built:
- One script in the toolchain check skill, with two modes.
- Write mode reads that run's
assumed: <ecosystem>.<key>=<value> (<file:line>)lines, re-reads each cited line, and writes the proposed record to a path the caller names outside the consumer's ecosystem records folder, so the same run never reads its own draft as a record and the unattended no-write rule holds. - Verify mode takes a record file and the lines and exits non-zero on any value the lines do not support.
- Write mode reads that run's
/toolchain:checkruns write mode only under an explicit flag.- The skill text tells the unattended session what to do with the draft: put it in the run report, and open a separate pull request that adds it to the ecosystem records folder through
/source-control:pull-request createwhen that skill is installed. The script neither commits nor pushes. No lane workflow file changes. - The skill section cites the acceptance sentence ("a merged pull request counts as acceptance", ecosystem records only).
Method:
- Tests first: three
assumed:lines yield exactly those three values; a fourth command not in the lines exits non-zero and writes nothing; a malformed line is refused and named; a line whose citedfile:linedoes not hold that manifest script, task-runner target or build entry point is refused; an install verb or a CI workflow file as the source is refused; the output validates against the ecosystem-commands schema. - The skill-text change runs the toolchain check cases as a two-version compare.
- Run budget: 120 runs (8 toolchain check cases plus 2 new, on the
opusandsonnetaliases, three runs per side). Select cases with--case, use--runs 3and a--max-cost-usdceiling, and start no batch after a usage-limit error. - Eval batches run in the one local eval-runner session in the main checkout. Open the pull request with the
do-not-mergelabel; the compare record releases it.
Decisions
- D5 (interview Q18, Q30): a lane's record pull request holds only stated or inferred commands, checked mechanically.
- Design Q40: one script is both the writer and the check; a model-written record could hold a command the run never executed.
- D8: the writer checks each value against its cited line, writes outside the records folder, has a verify mode, and the skill text (not a lane workflow) tells the session to open the record pull request (approved by the owner with the plan).
- Constraint: "a merged pull request counts as acceptance" applies to ecosystem records only.
Key interfaces
- Lane record input: lines of the form
assumed: <ecosystem>.<key>=<value> (<file:line>), produced by an unattended run. The record's own format is the existing ecosystem-commands schema. - The writer's command line: reads
assumed:lines on standard input, writes one record file, exit code. - "One script, run by
/toolchain:checkunder an explicit flag, builds a lane's proposed record only from that run'sassumed:lines and refuses any other command; it is both the writer and the check." - Two-way door: the lane record script.
Done when
- AC6: a lane's record pull request containing a command outside the unattended limits is refused by a deterministic check (the writer's refusal case exits non-zero).
- The script's test file passes, including every refusal case.
- The check skill's text names the
assumed:input and the separate pull request. - The security review gate passed; the compare record and this repository's dogfooding step are in the pull request body.
- AC7: nothing is cut here, so no control result is owed.
- AC10: no new enforcement-placing skill is added here.
Out of scope
- Any lane workflow file.
- Committing or pushing from the script.
Metadata
Metadata
Assignees
Labels
agent-readyFully specified and briefed; eligible for autonomous pickup from the frontier.Fully specified and briefed; eligible for autonomous pickup from the frontier.area: securitySecurity-relevant: vulnerability, hardening, or disclosure follow-up.Security-relevant: vulnerability, hardening, or disclosure follow-up.needs-triageNot yet classified. Floor until a type and one priority tier are set.Not yet classified. Floor until a type and one priority tier are set.
Type
Fields
Work status
ready
Priority
medium
Activity
github-actions commentedon Nov 13, 2026
Size: Medium
Complexity: Medium
Uncertainty: Known
Flags: none
Blocked by: #15733, #16033
Blast radius: one new script and its tests in the toolchain check skill, plus skill text and the
/toolchain:checkflag; read by lane runs and by consumers who run the verify mode in their own CI.Reversible: yes; the script is a two-way door and can be removed or reverted, and no lane workflow file changes.
Open questions: none
Understood: add one script with a write mode and a verify mode. Write mode builds a proposed ecosystem record only from a run's
assumed:lines. Any other command makes it exit non-zero and write nothing. The outcome, demo, refusal cases and done-when items are exact, and the design decisions are already recorded in the issue. The issue is a new capability, so it is typed as Feature. It is hardening work, not a report of an existing weakness, so it needs no security hold. The issue text lists the security review as its gate. The admission comment says blockers #15733 and #16033 are still open, and I list those two. I found no open duplicate and no merged pull request that already fixes this.Automated triage. A maintainer reviews it.
melodic-automation-lanes commentedon Nov 13, 2026
Not admitted to the issue lane yet: blocker #16033 is not closed by a listed human or a merge; blocker #15733 is not closed by a listed human or a merge.