agentleFS
sponge-b0bsponge-b0b/Polaris/.agents/skills/wiki-synthesize/SKILL.md

wiki-synthesize

Skill4 starsChanged 4 months agoApache-2.0View on GitHub

Scan the full Living Entity Wiki for recurring patterns across Rejected Approaches and Open Questions that no single entity states alone. Produces higher-inference hypotheses for human review and never mutates authoritative or wiki state.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

23 rules. 0 are common, 20 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 20

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Do not invoke it automatically from the $wiki-sync skill or the$wiki-lint skill, routine implementation, document creation, or ADR …

    line 30
  2. Distinctive:

    Run it when enough durable wiki knowledge has accumulated that cross-entity synthesis may reveal something useful.

    line 32
  3. Distinctive:

    Do not run it after every session.

    line 34
  4. Distinctive:

    Use Strict Invariants as contextual evidence where relevant.

    line 52
  5. Distinctive:

    Use Planned and Boundary Rationale only when they materially clarify the pattern.

    line 54
  6. Distinctive:

    Do not treat Routing Anchors, Categories, or implementation structure as synthesis evidence by themselves.

    line 56
  7. Distinctive:

    Prefer cross-entity evidence, but repeated causal evidence within one entity may be reported as an entity-local recurrence.

    line 62
  8. Distinctive:

    Do not report patterns based only on:

    line 64
  9. Distinctive:

    Look for Rejected Approaches that appear to fail for the same underlying architectural reason.

    line 75
  10. Distinctive:

    Do not mark an Open Question resolved merely because another entity provides relevant evidence.

    line 85
  11. Distinctive:

    If accumulated rejections/questions suggest that an assumption behind a Strict Invariant deserves review, frame it cautiously.

    line 89
  12. Distinctive:

    Every pattern must identify its supporting entries with:

    line 103
  13. Distinctive:

    Do not strengthen provenance during synthesis.

    line 110
  14. Very specific:

    If a supporting entry has an obvious health problem such as a broken citation or unresolved [source-conflict], note the limitation.

    line 120
  15. Very specific:

    consider the $to-adr-doc skill if a real architectural decision is needed.

    line 154
  16. Distinctive:

    consider a topology review if several entities may represent one boundary.

    line 155
  17. Distinctive:

    Do not perform the action automatically.

    line 157
  18. Distinctive:

    Add a Caveat when evidence quality or source health materially limits the conclusion.

    line 183
  19. Distinctive:

    If no meaningful pattern exists, report:

    line 185
  20. Distinctive:

    Do not lower the evidence threshold merely to produce output.

    line 191
  21. Distinctive:

    If the owner confirms a finding, use the appropriate normal lifecycle afterward:

    line 208
  22. Distinctive:

    Do not log the synthesis run itself.

    line 216
  23. Distinctive:

    create documents or ADRs;

    line 227

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Wiki Synthesize

$wiki-synthesize performs deliberate, higher-inference analysis across the full Living Entity Wiki.

It looks for:

  • recurring causes behind Rejected Approaches;
  • related Open Questions across entities;
  • repeated constraints that may indicate a broader architectural concern;
  • accumulated evidence that an assumption deserves review.

Its output is a signal for human judgment, never architectural authority.

Invocation

$wiki-synthesize is manual-only.

Retain:

  • disable-model-invocation: true for Claude Code;
  • the existing agents/openai.yaml manual invocation configuration for Codex.

Do not invoke it automatically from the $wiki-sync skill or the$wiki-lint skill, routine implementation, document creation, or ADR lifecycle.

Run it when enough durable wiki knowledge has accumulated that cross-entity synthesis may reveal something useful.

Do not run it after every session.

1. Load the Active Wiki

Read:

wiki/index.md
wiki/entities/*

Load every active entity page.

Primary synthesis inputs are:

  • Rejected Approaches;
  • Open Questions.

Use Strict Invariants as contextual evidence where relevant.

Use Planned and Boundary Rationale only when they materially clarify the pattern.

Do not treat Routing Anchors, Categories, or implementation structure as synthesis evidence by themselves.

2. Find Meaningful Patterns

A pattern requires at least two distinct supporting entries.

Prefer cross-entity evidence, but repeated causal evidence within one entity may be reported as an entity-local recurrence.

Do not report patterns based only on:

  • similar wording;
  • shared technologies;
  • two unrelated failures;
  • an abstraction the agent can merely imagine.

The causal reasoning must meaningfully overlap.

Recurring Rejections

Look for Rejected Approaches that appear to fail for the same underlying architectural reason.

Open-Question Relationships

Look for:

  • reinforcement — multiple entries independently raise the same concern;
  • partial answer — another entity contains relevant evidence;
  • tension — an unresolved question challenges an assumption elsewhere.

Do not mark an Open Question resolved merely because another entity provides relevant evidence.

Invariant Tension

If accumulated rejections/questions suggest that an assumption behind a Strict Invariant deserves review, frame it cautiously.

Valid:

Several entries suggest that the assumption behind invariant X may deserve review.

Invalid:

Invariant X is obsolete.

Direct contradictions belong to the $wiki-lint skill, not synthesis.

3. Preserve Provenance

Every pattern must identify its supporting entries with:

  • Entity ID/name;
  • section;
  • concise entry;
  • existing provenance.

Do not strengthen provenance during synthesis.

For example:

source: agent-observed during session, unresolved

must never become an owner decision merely because the pattern looks persuasive.

If a supporting entry has an obvious health problem such as a broken citation or unresolved [source-conflict], note the limitation.

Recommend the $wiki-lint skill when systematic source-health evaluation is required rather than duplicating lint behavior here.

4. Assign Confidence

Use:

Confidence: low | medium | high

Confidence means strength of synthesis evidence, not architectural authority.

  • Low — plausible but indirect or assumption-heavy.
  • Medium — multiple entries share a clear causal theme.
  • High — several independent entries strongly reinforce the same pattern with little interpretive stretching.

Even high means:

high confidence that the pattern deserves review.

It does not mean a decision has been established.

5. Suggest the Smallest Next Step

For each pattern, recommend only the smallest useful follow-up.

Examples:

  • review an Open Question;
  • reconsider a condition-dependent rejection;
  • inspect a Strict Invariant;
  • invoke the $wiki-lint skill if a direct conflict may exist;
  • consider the $to-adr-doc skill if a real architectural decision is needed;
  • consider a topology review if several entities may represent one boundary.

Do not perform the action automatically.

Report Format

Use:

## Synthesis Report — YYYY-MM-DD

### Pattern: [concise description]

**Confidence:** low | medium | high

**Interpretation:**
[Why the entries may share an underlying cause or concern.]

**Supporting entries:**
- [entity] › [section]: [concise entry]
  (source: ...)
- [entity] › [section]: [concise entry]
  (source: ...)

**Suggested next step:**
[Smallest appropriate human-reviewed follow-up.]

Add a Caveat when evidence quality or source health materially limits the conclusion.

If no meaningful pattern exists, report:

Synthesis: no meaningful recurring patterns found.

Do not lower the evidence threshold merely to produce output.

Report-Only Rule

$wiki-synthesize never writes to:

  • wiki/entities/;
  • wiki/index.md;
  • wiki/log.md;
  • docs/;
  • ADRs;
  • source code.

It never creates a commit.

The synthesis report itself is not an authoritative source.

If the owner confirms a finding, use the appropriate normal lifecycle afterward:

  • architectural decision → $to-adr-doc;
  • entity/wiki change → $wiki-sync;
  • direct conflict → $wiki-lint / source resolution;
  • new non-ADR document → $to-doc;
  • existing document reclassification → $classify-doc.

Do not log the synthesis run itself.

Out of Scope

$wiki-synthesize does not:

  • perform ordinary drift or citation auditing;
  • resolve [source-conflict];
  • maintain entity pages;
  • decide topology;
  • establish architectural authority;
  • create documents or ADRs;
  • convert inference directly into durable wiki knowledge.

Its job is to ask:

What meaningful pattern may be emerging across the knowledge already accumulated?

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.