feat: add /techne:research-grounded skill#31
Merged
Conversation
New /techne:research-grounded audits IMPL.md / ROADMAP.md for committed design
decisions (library / framework / pattern / architecture choices) that lack a
`# research(YYYY-MM):` provenance tag, then web-searches to ground them -- closing
the loop that, when skipped, turned an SSML capability bet into 5 revertable PRs.
Judgment over grep: the decision-language grep is only a candidate seed; the skill
reads around each hit and filters descriptive prose ("reads X instead of Y") and
hypotheticals ("what if we used..."). Validated on kourai's ROADMAP/IMPL via a fresh
subagent -- 8/10 grep candidates correctly ignored, 2 genuine gaps surfaced.
Wired into README, docs/skills/, and the zensical nav; moved Queued -> Shipped.
Sibling of /techne:docsync (claims vs code; this is decisions vs evidence).
ajbarea
added a commit
that referenced
this pull request
May 29, 2026
Adding /techne:research-grounded (#31) left "9 skills" claims behind in ROADMAP, IMPL, and skill-context — the count lives in more places than the slop_ground_truth note named. Bump all to 10, add research-grounded to the Drift dimension table and the skill-context name list, and widen slop_ground_truth to name every place a count claim lives (README, ROADMAP, IMPL, skill-context) so the next addition catches them all. The historical "eight of nine" spec note is left as-is (archival).
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
A new
/techne:research-groundedskill — the queued ROADMAP item that mechanizes the# research(YYYY-MM):provenance convention. It auditsIMPL.md/ROADMAP.mdfor committed design decisions (library / framework / pattern / architecture choices) that lack a research-provenance tag, then web-searches to ground them.The motivation, from the plan doc: when an SSML capability was assumed for Kourai's M6 and never verified against the target's docs (ElevenLabs v3 doesn't support SSML break tags), it became 5 PRs of revertable work. A decision stated as fact but never checked is the failure this skill catches — before it hardens into code.
Design: judgment over grep
The decision-language grep (
chose|switched|adopted|instead of|…) is only a candidate seed. Mostinstead ofhits are descriptive implementation prose ("reads X instead of re-reading Y"), not technology choices — so the skill reads around each hit and classifies per-hit. ItsIgnoresection explicitly waves through descriptive prose, hypotheticals ("what if we used…"), already-grounded decisions, and mechanical changes. This is why it's a skill (judgment), not a CI regex.Validation (GREEN test)
Ran the skill, report-only, via a fresh subagent against kourai's ROADMAP+IMPL (established as ground truth):
Message.metadataextension-key architecture choice; a word-level-timing pattern justified in prose but missing the formal tag).Low false-positive rate, real signal.
Registration
plugins/techne/skills/research-grounded/SKILL.mddocs/skills/research-grounded.md+ zensical nav entryLocal checks green:
validate_skill_frontmatter.py(all 10 skills ok), manifests jq-valid,zensical buildclean. Sibling of/techne:docsync— claims-vs-code there, decisions-vs-evidence here.